{ "status": "partial", "completed_count": 66, "applicable_count": 66, "failed_count": 0, "schema_version": 7, "config_hash": "dc07cd5433bf681a", "data": { "scenes": [ { "scene_index": 1, "heading": "S#1. 숲속 / 해 질 무렵", "beat_title": "공포에 질린 도주와 피 흘리는 상처", "representative_moment": "무언가에 물려 피가 흐르는 C07의 허벅지 클로즈업.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Extreme close-up (ECU) low angle at knee height, handheld shaking perspective, cold blue-gray dusk with sharp crimson accents", "t2i_prompt": "Photorealistic cinematic still. Extreme close-up low angle at knee height, tight framing on the bare thigh of a middle-aged East Asian man in torn fluorescent hiking pants. The leg is captured mid-stride stumbling forward through the frame. Focus on a large, jagged bite mark on the clammy skin occupying the center-left, with fresh sharp crimson blood drops tracking downward. [L01: Wet, overgrown forest brush scraping the blurred lower and side frame edges, muddy forest floor and soft dark vertical tree trunks in the deep background]. Cold blue-gray dusk light washes the edge of the leg, leaving the red blood as the sharpest point of contrast against the deep, pervasive shadows of the blurred foliage.", "outfit_assignments": [], "source_facts": [ "Scene text: 우거진 수풀을 헤치며 도망치는. 허벅지에 커다란 이빨 자국. 피가 흐르고.", "Shot 3 description: extreme close-up (ECU) low angle beside the bare thigh at knee height, bite mark center-left with blood drops tracking downward" ], "visual_inferences": [ "fixed_element dusk_forest_lighting: cold, fading blue-gray dusk light, deep pervasive shadows", "The action is frozen mid-stride to capture the panic of running while keeping the focus entirely on the wound", "Handheld motion feel translated into a frantic, tight composition with brush obscuring the edges" ], "creative_decisions": [ "Replaced character ID with common noun and demographic descriptor per body-part focus rule", "Applied shallow depth of field to heavily blur the background tree trunks and foreground brush, making the red blood the only readable detail", "Used cold blue-gray ambient light washing the skin to heighten the stark crimson contrast of the blood" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "d8fe24408709f47f", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "6aaacd9d20465357", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "N/A", "_shot_index": 3, "visible_entities": [ "C07", "P01", "L01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 1, "shot_index": 3 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "An observer's direct view holds an extreme close-up (ECU) low angle beside the 한국인 남자's bare thigh at knee height, handheld and shaking as brush scrapes across the frame edges. The bite mark sits center-left with blood drops tracking downward, while the rest of the body and forest are cropped into blurred fragments so the wound becomes the only readable point of panic.", "lighting_mood": "Cold blue-gray dusk light leaks through foliage, with the blood rendered as a sharp crimson accent against clammy skin.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C07", "P01", "L01" ], "allowed_outlook_pairs": [ { "character_id": "C07", "outlook_id": "O09" }, { "character_id": "C15", "outlook_id": "O15" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": null, "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "dusk_forest_lighting", "element_type": "environment_state", "character_name": "", "description": "A dense, overgrown forest environment enveloped in cold, fading blue-gray dusk light. The surrounding foliage and tree trunks are partially obscured by deep, pervasive shadows, creating a chilling and eerie atmosphere.", "applies_to_shots": [ 3, 4, 5 ], "source_facts": [ "Scene text: 숲속 / 해 질 무렵 (Forest / Dusk)", "Shot 3: Cold blue-gray dusk light leaks through foliage", "Shot 4: Fading coastal-blue dusk", "Shot 5: cold dusk tones, using the last gray sky" ], "visual_inferences": [ "Deep, pervasive shadows swallowing the background elements due to the extremely low light level" ], "creative_decisions": [ "Applying a consistent cold blue-gray color palette across extreme close-ups and wide shots to anchor the visual timeline right before nightfall" ], "confidence": "high" } ], "previous_shot_refs": [], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C07O09", "policy": "required" }, { "kind": "character_outlook", "id": "C15O15", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "98876b93167416ec" }, { "scene_index": 1, "heading": "S#1. 숲속 / 해 질 무렵", "beat_title": "알 수 없는 공포에 질린 얼굴", "representative_moment": "C07이 땀에 젖은 얼굴로 공포에 질려 어둠을 주시하며 비틀거리는 순간.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Eye-level close-up, handheld feel, shallow depth of field, cold blue-gray dusk lighting.", "t2i_prompt": "Photorealistic cinematic still. Tight framing on the expression of C07O09 in a neon hiking jacket, a middle-aged East Asian man, at three-quarter angle from the left, eye-level handheld perspective. His sweat-slick face shows breathless panic, pupils blown wide, eyes fixed on the unseen darkness past camera right. He is stumbling forward panting, his left shoulder visible at the lower frame edge. Lit by fading cold blue-gray dusk light catching damp skin highlights, against [L01: dense unreadable shadowed trees and thin blurred branches swaying near frame edges, deep pervasive shadows swallowing the narrow forest path]. Shallow depth of field blurring the background foliage into streaks, keeping his terrified face sharp.", "outfit_assignments": [ { "character_id": "C07", "outlook_id": "O09" } ], "source_facts": [ "Scene text: 잔뜩 겁에 질린 남자 표정", "shot description: sweat-slick face, pupils blown wide, eye-level handheld shot from his front-left" ], "visual_inferences": [ "fixed_element dusk_forest_lighting: fading cold blue-gray dusk light, deep pervasive shadows", "shoulder visible at the lower frame edge to avoid floating head synthesis" ], "creative_decisions": [ "tight head-and-shoulder framing instead of full-frame face to ensure proper body attachment synthesis", "handheld camera blur on background trees to emphasize frantic movement" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "86ae97633ca9ed51", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "0d9873c7e50d66b8", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 1, "dependency_reason": "이전 샷의 숲속 해 질 무렵 조명 톤 유지 (close framing으로 배경은 엣지만 새로 구성)", "_shot_index": 4, "visible_entities": [ "C07", "P01", "L01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 1, "shot_index": 4 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "An observer's direct view shifts to a face-filling close-up (CU) eye-level handheld shot from the 한국인 남자's front-left, backing away just ahead of him on the forest path. His sweat-slick face occupies nearly the whole frame, pupils blown wide and gaze snapping past camera right toward the unseen darkness, with shallow DOF turning the trees into streaks.", "lighting_mood": "Fading coastal-blue dusk and damp skin highlights create a cold, breathless panic, with shadows swallowing the background.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C07", "P01", "L01" ], "allowed_outlook_pairs": [ { "character_id": "C07", "outlook_id": "O09" }, { "character_id": "C15", "outlook_id": "O15" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": null, "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "dusk_forest_lighting", "element_type": "environment_state", "character_name": "", "description": "A dense, overgrown forest environment enveloped in cold, fading blue-gray dusk light. The surrounding foliage and tree trunks are partially obscured by deep, pervasive shadows, creating a chilling and eerie atmosphere.", "applies_to_shots": [ 3, 4, 5 ], "source_facts": [ "Scene text: 숲속 / 해 질 무렵 (Forest / Dusk)", "Shot 3: Cold blue-gray dusk light leaks through foliage", "Shot 4: Fading coastal-blue dusk", "Shot 5: cold dusk tones, using the last gray sky" ], "visual_inferences": [ "Deep, pervasive shadows swallowing the background elements due to the extremely low light level" ], "creative_decisions": [ "Applying a consistent cold blue-gray color palette across extreme close-ups and wide shots to anchor the visual timeline right before nightfall" ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 1, "shot_index": 3, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C07O09", "policy": "required" }, { "kind": "character_outlook", "id": "C15O15", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "fb5b08f9419d0736" }, { "scene_index": 1, "heading": "S#1. 숲속 / 해 질 무렵", "beat_title": "공포에 질려 도망치던 남자가 검은 형상에게 덮쳐지는 순간", "representative_moment": "C15O15 leaps from the dark canopy, mid-air with arms outstretched, lunging toward C07O09 who recoils in horror below.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Observer's low-angle perspective looking upward, underexposed cold dusk tones with silhouette backlighting", "t2i_prompt": "Photorealistic cinematic still. Observer's view looking upward from the dark forest floor. C15O15 in a tattered dark cloak, an East Asian figure, occupies the upper center of the frame in mid-leap, facing the camera with arms outstretched, cloak trailing back in the direction opposite to motion. C15O15 appears as a backlit attacker against the dim sky, rim light highlighting the shoulders and jawline, body volume preserved against the dim ambient fill, face in deep shadow but contour identifiable, eyes fixed on the man below. C07O09 in a fluorescent hiking jacket, a middle-aged East Asian man, occupies the lower foreground with back to camera, legs and torso visible, recoiling on his back foot, head turned slightly to reveal his side profile, eyes fixed on C15O15 above. P01, a worn shoe, lies on the muddy ground near his feet. Both figures share the same muddy, uneven clearing ground covered with scattered leaves. Focus on the airborne C15O15 dominating the upper frame. The scene is underexposed in cold, fading blue-gray dusk light. The dense, shadowed tree canopy forms a dark border around the upper frame, framing the clouded gray-blue evening sky acting as a backlight.", "outfit_assignments": [ { "character_id": "C15", "outlook_id": "O15" }, { "character_id": "C07", "outlook_id": "O09" } ], "source_facts": [ "Scene text: 우거진 수풀을 헤치며 도망치는. 벗겨진 한쪽 신발. 어디선가 쏜살같이 달려온 검은 인영이 남자를 덮친다.", "Shot 5 description: 짙은 어둠 속에서 나타난 검은 형상이 한국인 남자를 덮치기 위해 공중으로 뛰어올라 양팔을 뻗은 채 허공에 떠 있는 찰나의 순간." ], "visual_inferences": [ "fixed_element dusk_forest_lighting: A dense, overgrown forest environment enveloped in cold, fading blue-gray dusk light.", "C15O15's cloak trails back to preserve motion direction during the mid-leap freeze.", "C07O09's head is turned slightly to make the facial contour identifiable from behind, complying with OTS and silhouette ID rules." ], "creative_decisions": [ "Positioned C15O15 in the upper center and C07O09 in the lower foreground to naturally create a low-angle observer's perspective without using banned shot type keywords.", "Added rim light highlighting C15O15's shoulders and jawline to prevent a flat cutout silhouette, maintaining body volume against the dim sky.", "Explicitly established the shared muddy ground anchor to prevent foreground/background separation errors." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "1ccbb5a34f8d081f", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "d7231eb8d11090b0", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 1, "dependency_reason": "Shot 4의 배경(해 질 무렵의 숲속, 차가운 푸른빛의 일몰)을 일관되게 유지하기 위함", "_shot_index": 5, "visible_entities": [ "C07", "C15", "P01", "L01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 1, "shot_index": 5 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "An observer's direct view locks into a wide shot (WS) low angle from the dark clearing, static and aimed upward as 검은 형상 bursts out of the upper center of frame. The 한국인 남자 stays small and lower in the frame, recoiling beneath it, while the black figure is held at the apex of its leap as a silhouette against the dim sky.", "lighting_mood": "The scene is underexposed in cold dusk tones, using the last gray sky as backlight to carve the attacker into a black occult silhouette.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C07", "C15", "P01", "L01" ], "allowed_outlook_pairs": [ { "character_id": "C07", "outlook_id": "O09" }, { "character_id": "C15", "outlook_id": "O15" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "not_applicable", "bg_id": null, "reference_usage": "none", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "no chain_bg is bound to this shot — describe the location freely without referencing a background image", "do not use 'from the reference' / 'the existing room' / 'preserving the room' style phrasing — there is no reference image to refer to" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "dusk_forest_lighting", "element_type": "environment_state", "character_name": "", "description": "A dense, overgrown forest environment enveloped in cold, fading blue-gray dusk light. The surrounding foliage and tree trunks are partially obscured by deep, pervasive shadows, creating a chilling and eerie atmosphere.", "applies_to_shots": [ 3, 4, 5 ], "source_facts": [ "Scene text: 숲속 / 해 질 무렵 (Forest / Dusk)", "Shot 3: Cold blue-gray dusk light leaks through foliage", "Shot 4: Fading coastal-blue dusk", "Shot 5: cold dusk tones, using the last gray sky" ], "visual_inferences": [ "Deep, pervasive shadows swallowing the background elements due to the extremely low light level" ], "creative_decisions": [ "Applying a consistent cold blue-gray color palette across extreme close-ups and wide shots to anchor the visual timeline right before nightfall" ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 1, "shot_index": 4, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C07O09", "policy": "required" }, { "kind": "character_outlook", "id": "C15O15", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "9d062b0ddd86aa82" }, { "scene_index": 2, "heading": "S#2. 도시 뒷골목 / N", "beat_title": "남자의 이빨이 여자의 목에 닿는 정지 순간", "representative_moment": "Extreme close-up on the jaw of an East Asian man pressing against the pale neck of an East Asian woman.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Extreme close-up (ECU) eye-level side profile at neck height", "t2i_prompt": "Photorealistic cinematic still. Extreme close-up on the pale throat of a young East Asian woman and the jaw of a young East Asian man. The woman is in left profile, her head tilted back, her eyes looking up out of frame, the collar of her top visible at the lower edge. The man is in right profile, wearing a dark hood, his jaw pressed against her neck, his eyes fixed on her throat, his teeth just touching the bloodless skin at dead center. Sickly sodium-green light grazes their cold skin, highlighting the stark contact. [L03: background dissolved into abstract yellow-green streetlight bokeh and soft vertical darkness]. The frame is unnaturally still.", "outfit_assignments": [], "source_facts": [ "Scene text: '남자의 붉은 눈동자, 목의 살점을 뜯는 이빨…'", "Shot 4 description: observer's direct view freezes into an extreme close-up (ECU) eye-level side profile at neck height, only inches from the pale throat and mouth. The teeth touch skin at dead center." ], "visual_inferences": [ "fixed_element wet_alleyway_night environment translates to abstract streetlight bokeh due to skipped_close_framing policy.", "Body part focus rule applied: replaced C##O## with common nouns and demographic descriptors to prevent floating head artifacts in extreme close-up framing." ], "creative_decisions": [ "Frame is locked tightly on the contact point between the jaw and neck to maximize tension, anchoring the lower edge with the collar of the top.", "Sickly sodium-green light pools on the skin surface to create a cold, corpse-like atmosphere without slicing the anatomy." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "9853fd93e184ae77", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "e80ec516404894be", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "Not directly dependent on a prior visual setup for framing", "_shot_index": 4, "visible_entities": [ "C08", "L03" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 2, "shot_index": 4 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "An observer's direct view freezes into an extreme close-up (ECU) eye-level side profile at neck height, only inches from the 한국인 여자's pale throat and the 한국인 남자's mouth. The teeth touch skin at dead center, with the background dissolved into abstract streetlight bokeh and the frame held almost unnaturally still.", "lighting_mood": "Sickly sodium-green alley light grazes the skin, making the bloodless neck cold and corpse-like before the crimson break.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C08", "L03" ], "allowed_outlook_pairs": [ { "character_id": "C08", "outlook_id": "O10" }, { "character_id": "C09", "outlook_id": "O11" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": null, "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "wet_alleyway_night", "element_type": "environment_state", "character_name": "", "description": "A cold, dark city alleyway at night. The space is illuminated by weak, sickly sodium-green and muted fluorescent streetlights, casting a gloomy and chilling atmosphere. The ground consists of wet, dark pavement reflecting the dim urban light.", "applies_to_shots": [ 4, 5 ], "source_facts": [ "Scene text: '도시 뒷골목 / N', '도심 희미한 가로등 불빛'", "Shot 4: 'Sickly sodium-green alley light grazes the skin'", "Shot 5: 'wet pavement and trash-lined walls', 'Muted blue-black night tones mix with weak fluorescent street spill'" ], "visual_inferences": [ "The overall dampness and wet pavement persist throughout the scene, even when blurred in the close-up shot." ], "creative_decisions": [ "Anchoring the color palette to a mix of sickly sodium-green and muted blue-black to unify the extreme close-up and the wide shot under the same eerie urban night atmosphere." ], "confidence": "high" } ], "previous_shot_refs": [], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C08O10", "policy": "required" }, { "kind": "character_outlook", "id": "C09O11", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "720ec68ad96204f9" }, { "scene_index": 2, "heading": "S#2. 도시 뒷골목 / N", "beat_title": "목을 물려 쓰러진 여자", "representative_moment": "어둡고 젖은 골목길 바닥에 목에 피를 흘린 C09가 쓰러져 있다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Wide shot from a low ground-level observer angle, featuring muted blue-black night tones and weak fluorescent street spill.", "t2i_prompt": "Photorealistic cinematic still. Low observer at ground level, the wet, dark pavement occupying the lower frame. C09O11 in a white blouse, a young Asian woman, lying motionless on the floor at a three-quarter angle from the right, with eyes closed. Her full body rests across the center of the cold pavement, arms limp. A sharp red wound is visible on her neck, drawing visual focus. [L03: A cold, dark city alleyway at night, enclosed by stained damp concrete walls. Discarded trash bags and damp cardboard line the frame edges. Muted blue-black night tones mix with weak, sickly sodium-green fluorescent street spill from a faint streetlight facing down the alley, reflecting on the wet ground]. Focus on the motionless figure anchored by the stark crimson contrast at her neck.", "outfit_assignments": [ { "character_id": "C09", "outlook_id": "O11" } ], "source_facts": [ "Scene text: '푹 힘없이 쓰러지는 여자. 도심 희미한 가로등 불빛.'", "Shot 5 description: 목에 피를 흘린 한국인 여자가 눈을 감은 채 차가운 뒷골목 바닥에 쓰러져 멈춘 전신 구도." ], "visual_inferences": [ "fixed_element wet_alleyway_night: The space is illuminated by weak, sickly sodium-green and muted fluorescent streetlights, casting a gloomy and chilling atmosphere on wet pavement.", "C09 is positioned in the center of the alley, forming a stark visual contrast against the dark pavement." ], "creative_decisions": [ "Used a low ground-level framing to push the concrete walls into the background and let the wet pavement dominate the lower frame, anchoring the body.", "Included discarded trash bags and damp cardboard at the edges to provide gritty urban scale without cluttering the center frame." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "e6bb946cffa3362c", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "9b5c8fad97d94811", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 2, "dependency_reason": "동일 씬 내 이전 샷들의 어둡고 축축한 골목길 조명(sickly sodium-green) 및 공간적 연속성 유지", "_shot_index": 5, "visible_entities": [ "C09", "L03" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 2, "shot_index": 5 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "An observer's direct view pulls back into a wide shot (WS) low angle from ground level, dollying away until the 한국인 여자's full body lies across the center of the alley floor. Her neck wound remains the visual anchor near the upper third of her body, while wet pavement and trash-lined walls stretch around her as cold negative space.", "lighting_mood": "Muted blue-black night tones mix with weak fluorescent street spill, leaving the blood at her neck as the only sharp red note.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C09", "L03" ], "allowed_outlook_pairs": [ { "character_id": "C08", "outlook_id": "O10" }, { "character_id": "C09", "outlook_id": "O11" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "not_applicable", "bg_id": null, "reference_usage": "none", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "no chain_bg is bound to this shot — describe the location freely without referencing a background image", "do not use 'from the reference' / 'the existing room' / 'preserving the room' style phrasing — there is no reference image to refer to" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "wet_alleyway_night", "element_type": "environment_state", "character_name": "", "description": "A cold, dark city alleyway at night. The space is illuminated by weak, sickly sodium-green and muted fluorescent streetlights, casting a gloomy and chilling atmosphere. The ground consists of wet, dark pavement reflecting the dim urban light.", "applies_to_shots": [ 4, 5 ], "source_facts": [ "Scene text: '도시 뒷골목 / N', '도심 희미한 가로등 불빛'", "Shot 4: 'Sickly sodium-green alley light grazes the skin'", "Shot 5: 'wet pavement and trash-lined walls', 'Muted blue-black night tones mix with weak fluorescent street spill'" ], "visual_inferences": [ "The overall dampness and wet pavement persist throughout the scene, even when blurred in the close-up shot." ], "creative_decisions": [ "Anchoring the color palette to a mix of sickly sodium-green and muted blue-black to unify the extreme close-up and the wide shot under the same eerie urban night atmosphere." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 2, "shot_index": 4, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C08O10", "policy": "required" }, { "kind": "character_outlook", "id": "C09O11", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "e886df3821db8f2c" }, { "scene_index": 3, "heading": "S#3. 현장 / D", "beat_title": "사건 현장을 지휘하며 바닥의 혈흔을 주시하는 혜수", "representative_moment": "C02 rests her chin on her fist, staring at a dark red bloodstain on the hazy forest floor.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Medium close-up (MCU) angled slightly down, flat overcast daylight with muted grays and subtle digital news texture", "t2i_prompt": "Photorealistic cinematic still. Medium close-up, angled slightly down. C02O03 in a gray leather jacket, a young East Asian woman, occupies the right middle ground at a three-quarter angle from the right. She rests her chin on her fist, her eyes fixed downward. Focus on the spatial relationship between her face and the irregular dark red-black bloodstain anchoring the lower left frame on [L02: a damp, muddy forest floor]. Flat overcast daylight filtered through dense fog washes over her, emphasizing muted grays and desaturated greens with a subtle digital news camera texture. Softened, out-of-focus silhouettes of moving figures and a sagging yellow police line are faintly visible in the soft background.", "outfit_assignments": [ { "character_id": "C02", "outlook_id": "O03" } ], "source_facts": [ "scene text: 혜수의 모습이 보인다", "shot description: 흐린 빛 속에서 혜수가 턱을 괸 채 바닥의 검붉은 핏자국을 응시하는 정지 구도", "DP instruction: broadcast-news observer view holds a medium close-up (MCU) high angle from low beside 혜수, angled slightly down" ], "visual_inferences": [ "A subtle digital texture is applied to reflect the 'through_device' broadcast news perception mode.", "Forensic officers and the police line are pushed to the out-of-focus background to maintain the close framing on the primary subject." ], "creative_decisions": [ "Placed C02O03 on the right and the bloodstain on the lower left to create a balanced diagonal anchor in the frame.", "Used desaturated greens and muted grays to reinforce the cold, forensic stillness of the modern Korean occult horror setting." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "1d65d924978ebd16", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "aebd9b965ca5cfa2", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "No prior scene visually binds the exact composition of this specific news broadcast shot.", "_shot_index": 3, "visible_entities": [ "C02", "P02", "P03", "L02" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 3, "shot_index": 3 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "A broadcast-news observer view holds a medium close-up (MCU) high angle from low beside 혜수, locked off about one meter away and angled slightly down so her chin-resting profile and the bloodstain share the frame. 혜수 occupies the right middle ground, the dark red-black stain anchors the lower left, and hazy police activity remains softened behind the line.", "lighting_mood": "Flat overcast daylight is filtered through fog, with muted grays and desaturated greens giving the news image a cold, forensic stillness.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C02", "P02", "P03", "L02" ], "allowed_outlook_pairs": [ { "character_id": "C02", "outlook_id": "O03" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": null, "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C02O03", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "db8a52a656f4bcd1" }, { "scene_index": 4, "heading": "S#4. 옥탑방 / D", "beat_title": "불안감이 감도는 텅 빈 옥탑방 전경", "representative_moment": "흐린 회색 하늘 아래 파란 물탱크가 돋보이는 낡은 옥탑방의 정적이고 고립된 전경.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "high oblique aerial view, diffuse gray daylight, muted blues and concrete grays", "t2i_prompt": "Photorealistic cinematic still. Observed from a high oblique aerial position looking down at the urban architecture. Focus on the worn rooftop room and the faded blue water tank from the reference, positioned slightly right of center in the frame. The roof terrace and parapet wall from the reference occupy the lower section. Neighboring villa rooftops clutter the outer edges, while a heavy overcast gray sky fills the top space. Diffuse gray daylight flattens the surfaces into muted blues and concrete grays, creating a tense, isolated atmosphere without any visible figures.", "outfit_assignments": [], "source_facts": [ "S#4 action: '인천 변두리 부감, 오래된 빌라 옥탑방 전경.'", "shot description: high oblique aerial position above the Incheon villa rooftops, old tile roof, blue water tank", "scene_text: only voiceover present, no physical characters in frame" ], "visual_inferences": [ "Empty frame emphasizes the VoiceOver's tension regarding the unknown suspect", "The water tank and parapet wall act as physical anchors for the bleak urban setting" ], "creative_decisions": [ "Removed framing scale keywords (EWS, wide shot) from the prompt, replacing them with spatial arrangement descriptions (from a high oblique aerial position looking down)", "Anchored background elements using 'from the reference' to comply with the background_ref_attached constraints" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "a6217b68a631ce28", "owned_hash": "c9ee8c218343e53d", "camera_direction_hash": "84e6d0b7360fb128", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "water tank", "violating_phrase": "the faded blue water tank from the reference", "reason": "The phrase describes the water tank as a scene element to be shown, which implies re-rendering an owned background object rather than merely referencing it." }, { "owned_object": "roof terrace", "violating_phrase": "The roof terrace ... from the reference occupy the lower section", "reason": "This explicitly places the roof terrace as part of the composition, which is a request to depict an owned background object again." }, { "owned_object": "parapet wall", "violating_phrase": "the parapet wall from the reference occupy the lower section", "reason": "This explicitly places the parapet wall as part of the composition, which is a request to depict an owned background object again." } ] } } ], "scene_type": "voiceover", "dependent_scene_index": -1, "dependency_reason": "V.O.가 흐르는 전경(Establishing) 샷으로 이전 인물 씬과의 직접적인 시공간적 연속성이 없음", "_shot_index": 1, "visible_entities": [ "L04" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 4, "shot_index": 1 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "An omniscient direct view settles into an extreme wide shot (EWS) high angle from a high oblique aerial position above the Incheon villa rooftops. The worn rooftop room sits slightly right of center, its old tile roof and blue water tank held in a static frame with enough surrounding gray sky and neighboring rooflines to make it feel isolated.", "lighting_mood": "Diffuse gray daylight flattens the buildings into muted blues and concrete grays, making the water tank's faded blue feel strangely ominous.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "L04" ], "allowed_outlook_pairs": [], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "bg_rooftop_day_overcast_establishing", "reference_usage": "exact_background", "owned_objects": [ "access door", "laundry line", "parapet wall", "post", "roof drain", "roof terrace", "sliding window", "stairwell", "wall light", "water tank", "window sill" ], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "owned_objects lists all environment objects already drawn in the background reference — do not create new ones", "when focusing on an owned object, anchor it explicitly: 'the door from the reference'" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "background", "id": "bg_rooftop_day_overcast_establishing", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "de06949218c8423d", "status": "contract_violation" }, { "scene_index": 5, "heading": "S#5. 옥탑방 안 / 실내", "beat_title": "TV 뉴스를 보며 굳어있는 민숙", "representative_moment": "부엌에 선 민숙(C05)이 어두운 표정으로 텔레비전 뉴스를 뚫어지게 바라보고 있다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Close-up eye-level frontal portrait, mixed pale daylight and sickly TV-blue spill, shallow depth of field", "t2i_prompt": "Photorealistic cinematic still. A tight eye-level frontal portrait of C05O06 in a beige cardigan and apron, a middle-aged East Asian woman, facing the camera. Head-and-shoulder composition, her anxious face occupying the center frame, shoulders tight beneath the apron visible in the lower frame, hands gripping the apron bib entering the bottom edge. Her eyes are fixed intensely off-camera toward an unseen TV. A sickly, cool blue spill from the television washes across her cheek and jawline, mixing with pale daylight reflecting on her opposite shoulder. Behind her, [L07: the shallow background is minimized into a soft, out-of-focus domestic haze, thick misty steam rising from the lower edge and obscuring the dim pale light of a small window]. Focus on her rigid, tense expression.", "outfit_assignments": [ { "character_id": "C05", "outlook_id": "O06" } ], "source_facts": [ "S#5 scene text: 앞치마를 두른 채 TV 뉴스를 뚫어지게 바라보는 민숙(수리영 엄마. 40세) 표정이 어둡다.", "Shot 4 description: An observer's direct view pushes into a close-up (CU) eye-level frontal portrait... Her face fills most of the center frame, shoulders tight beneath the apron" ], "visual_inferences": [ "fixed_element steamy_kitchen_background: soft misty haze of steam rising behind her", "hands gripping the apron bib to naturally incorporate the DP's 'gripping apron hem' within a close head-and-shoulder composition" ], "creative_decisions": [ "mixed lighting (sickly TV-blue spill and pale window daylight) washing across her face to heighten the tense, anxious atmosphere", "head-and-shoulder composition used to comply with close-framing policy and avoid floating-head artifacts", "shallow depth of field to soften the background and keep the focus entirely on her expression" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "265b4ecb4b4dcc21", "owned_hash": "cb1d462ffc515e90", "camera_direction_hash": "23afcc72051db611", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "N/A", "_shot_index": 4, "visible_entities": [ "C05", "P04", "L07" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 5, "shot_index": 4 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "An observer's direct view pushes into a close-up (CU) eye-level frontal portrait of 40대 한국인 여성 from the living-room side, keeping the TV just off her eyeline rather than visible head-on. Her face fills most of the center frame, shoulders tight beneath the apron, while the sink steam and small window are minimized into soft domestic haze behind her.", "lighting_mood": "Pale window light mixes with sickly TV-blue spill, draining her face into anxious gray while the kitchen steam softens the room.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C05", "P04", "L07" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C05", "outlook_id": "O06" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_living_kitchen_day_normal", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "steamy_kitchen_background", "element_type": "environment_state", "character_name": "", "description": "The kitchen sink area is enveloped in a soft, misty haze of steam rising from a recently boiled pot. The surrounding space is dimly lit by pale natural light from a small window, creating a muted, slightly hazy domestic atmosphere in the background.", "applies_to_shots": [ 4, 9 ], "source_facts": [ "S#5 scene text: 냄비의 물이 끓어 달그락거리는 소리… 김이 피어오르고", "Shot 4 description: sink steam and small window are minimized into soft domestic haze behind her", "Shot 9 description: blurred sink and shut-off TV behind her... leftover kitchen steam create a muted domestic calm" ], "visual_inferences": [ "The steam from the boiling medicinal herbs consistently permeates the kitchen area, visibly affecting the background clarity and atmosphere." ], "creative_decisions": [ "Using the lingering steam as a continuous environmental texture to spatially connect the separate close-up shots of the mother and daughter." ], "confidence": "high" } ], "previous_shot_refs": [], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O01", "policy": "required" }, { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C05O06", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "2cec06233eb02d92" }, { "scene_index": 5, "heading": "S#5. 옥탑방 안 / 실내", "beat_title": "맛없는 약초물과 심드렁한 아침", "representative_moment": "수리영(C01)이 컵을 입술에 댄 채 미간을 찌푸린 얼굴 클로즈업.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "eye-level close-up, three-quarter angle, shallow depth of field with soft gray daylight", "t2i_prompt": "Photorealistic cinematic still. Eye-level head-and-shoulder composition. C01O01 in a gray homewear top, a young East Asian woman, seated at the dining table at a three-quarter angle from the left. A cup filled with medicinal brown-green herbal water is held in her hand, the rim angled toward the camera and resting against her lips in the lower center of the frame. Focus on her slight frown and narrowed eyes fixed on the cup, with her neck and shoulders visible in the frame. Behind her, [L07: a softly blurred domestic space enveloped in a misty haze of steam, featuring the wet faucet side of a kitchen sink, a dark shut-off TV screen angled away, and pale daylight]. Soft gray natural light catches on her face and the cup, creating a muted, slightly hazy domestic calm.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "S#5 scene text: 냄비의 물이 끓어 달그락거리는 소리… 김이 피어오르고", "Shot 9 description: 컵을 입술에 댄 채 미간을 살짝 찌푸린 수리영의 얼굴 클로즈업", "DP instruction: The cup rests against her lips in the lower center of frame... slight frown and narrowed eyes hold focus against a blurred sink and shut-off TV behind her." ], "visual_inferences": [ "fixed_element steamy_kitchen_background: The kitchen sink area is enveloped in a soft, misty haze of steam", "C01 wearing homewear (O01) as she just woke up and came out of her room." ], "creative_decisions": [ "Used a shallow depth of field to keep the focus tight on her facial expression while soft-blurring the kitchen background, adhering to the skipped_close_framing background rule.", "Anchored the cup at her lips in the lower center to draw the viewer's eye to her displeased gaze fixed on the medicinal water." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "e0f535274e15185a", "owned_hash": "cb1d462ffc515e90", "camera_direction_hash": "8dc827d854edf684", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 5, "dependency_reason": "previous_shot_refs specifies S5_Shot4 for continuity of the steamy kitchen background, though handled here as a softly blurred frame-edge surface due to close framing rules.", "_shot_index": 9, "visible_entities": [ "C01", "P04", "L07" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 5, "shot_index": 9 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "An observer's direct view resolves the table dolly into a close-up (CU) eye-level three-quarter angle on 수리영 at the dining table. The cup rests against her lips in the lower center of frame, taking less than a third of the image, while her slight frown and narrowed eyes hold focus against a blurred sink and shut-off TV behind her.", "lighting_mood": "Soft gray daylight and leftover kitchen steam create a muted domestic calm, with the herbal water tinted a faint medicinal brown-green.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "P04", "L07" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C05", "outlook_id": "O06" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_living_kitchen_day_normal", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "steamy_kitchen_background", "element_type": "environment_state", "character_name": "", "description": "The kitchen sink area is enveloped in a soft, misty haze of steam rising from a recently boiled pot. The surrounding space is dimly lit by pale natural light from a small window, creating a muted, slightly hazy domestic atmosphere in the background.", "applies_to_shots": [ 4, 9 ], "source_facts": [ "S#5 scene text: 냄비의 물이 끓어 달그락거리는 소리… 김이 피어오르고", "Shot 4 description: sink steam and small window are minimized into soft domestic haze behind her", "Shot 9 description: blurred sink and shut-off TV behind her... leftover kitchen steam create a muted domestic calm" ], "visual_inferences": [ "The steam from the boiling medicinal herbs consistently permeates the kitchen area, visibly affecting the background clarity and atmosphere." ], "creative_decisions": [ "Using the lingering steam as a continuous environmental texture to spatially connect the separate close-up shots of the mother and daughter." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 5, "shot_index": 4, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O01", "policy": "required" }, { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C05O06", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "29ed7ffd61db873a" }, { "scene_index": 5, "heading": "S#5. 옥탑방 안 / 실내", "beat_title": "손목의 붉은 표식 인지", "representative_moment": "민숙의 손목에 있는 작은 원형 표식이 붉은빛을 띠며 드러난 순간.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "High-angle view directed at the wrist, dim bluish ambient shadow contrasting with a faint vivid red glow", "t2i_prompt": "Photorealistic cinematic still. High-angle view at waist height, looking down directly onto the bare wrist of a middle-aged East Asian woman, focusing tightly on a small circular mark on the skin. The mark sits just right of center, holding a faint red glow that feels vividly unnatural against the flesh. The edge of a dark sleeve is pushed back slightly, framing the wrist. Dim indoor daylight washes across the arm before falling off into deep bluish shadow. In the soft, out-of-focus background plane beneath the hand, [L07: a worn entryway floor and the blurred edge of a shoe rack recede into the dim, cool lighting].", "outfit_assignments": [], "source_facts": [ "S#5 action: '힐끗, 민숙 손목에 보이는 작은 원(이하 표식)을 본다. 오늘따라 유난히 또렷해 보이지만...'", "shot description: A close-up of a small circular mark with a faint red glow on a Korean woman's bare wrist." ], "visual_inferences": [ "Sleeve pushed back slightly to reveal the bare wrist naturally.", "A downward perspective emphasizing the act of noticing the mark from above.", "Dim entryway lighting with cool shadows enhancing the supernatural feel of the red mark." ], "creative_decisions": [ "Used a common noun with demographic descriptor instead of entity ID due to the body-part focus rule.", "Omitted full body descriptions to maintain strict focus on the wrist.", "Incorporated L07 description strictly for the out-of-focus background surface to anchor the space without relying on skipped background references." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "a8bf9123675cb2d6", "owned_hash": "cb1d462ffc515e90", "camera_direction_hash": "d02853eff15fec09", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 5, "dependency_reason": "S5_Shot9의 공간 설정 및 조명 톤을 바탕으로, 부분 포커스 구도에 맞게 배경(바닥) 블러 처리 연계", "_shot_index": 14, "visible_entities": [ "P04", "L07" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 5, "shot_index": 14 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "An observer's direct view drops into a close-up (CU) high angle from beside 40대 한국인 여성 at chest-to-waist height, angled down onto her bare wrist as the dolly-in settles. The small circular mark sits just right of center on the skin with a faint red glow, while the edge of her sleeve and the entryway floor remain visible as scale references.", "lighting_mood": "Dim indoor daylight falls off into bluish entryway shadow, making the mark's crimson tint feel unnaturally vivid against ordinary skin.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "P04", "L07" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C05", "outlook_id": "O06" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_living_kitchen_day_normal", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [ { "scene_index": 5, "shot_index": 9, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O01", "policy": "required" }, { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C05O06", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "7950d7d661b86493" }, { "scene_index": 6, "heading": "S#6. 해안가 도로 / D", "beat_title": "해안가 도로를 달리는 수리영의 평온한 옆얼굴", "representative_moment": "하얀 이어폰을 꽂고 눈을 감은 채 자전거를 타는 C01의 평온한 옆얼굴 클로즈업.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Tight eye-level head-and-shoulder close-up, parallel tracking, muted blue-gray cold coastal daylight", "t2i_prompt": "Photorealistic cinematic still. A tight eye-level head-and-shoulder composition. C01O02 in a navy windbreaker, a young East Asian woman, is framed in left profile, occupying the right half of the image. She leans slightly forward into her ride, her expression premonitory and calm with eyes closed. A crisp white earphone rests securely in her visible ear. Cold coastal daylight diffused through thick sea fog washes her cheek and the jacket collar in a pale blue-gray softness. Behind her, [L08: the coastal road and hazy sea smear into a muted blue-gray bokeh]. At the lower frame edge, a sliver of the bicycle handlebar shows a faint motion blur, implying forward movement against the dense fog.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "S#6 action: '수리영 자전거를 타고 해안가 도로를 달린다. 이어폰을 끼고 바다냄새를 실컷 마신다.'", "shot description: 양 귀에 하얀 이어폰을 꽂고 지긋이 눈을 감은 수리영의 옆얼굴 클로즈업", "character pose: profile_left, leaning into ride, eyes closed" ], "visual_inferences": [ "head-and-shoulder framing preserves the upper chest and jacket collar, preventing a floating-face artifact", "motion blur on the handlebar at the bottom edge implies bicycle movement consistently with the close framing constraint", "muted blue-gray bokeh for the background applies the skipped_close_framing logic by relying on immediate soft surroundings" ], "creative_decisions": [ "subject placed in the right half of the frame in left profile to provide look-room in the direction of the implied movement", "cold coastal daylight directly washing the skin surface to establish the required premonitory and quiet mood" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "23005c1901ee4e4f", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "9c82441cf3ab8c61", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "해당 없음", "_shot_index": 2, "visible_entities": [ "C01", "L08" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 6, "shot_index": 2 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Observer perspective, direct vision. A tight eye-level CU tracks parallel about two feet from 수리영's left side, holding her calm side-profile in the right half of frame while both white earphones and her closed eyelids stay crisp; the bicycle motion is implied only by a faint tremor at the frame edge. The foggy coastal road and sea smear into muted blue-gray bokeh behind her, with no independent camera shift beyond the flow’s close side observation.", "lighting_mood": "Cold coastal daylight is diffused through sea fog, washing her skin in pale blue-gray softness with a quiet, almost premonitory calm.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "L08" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": null, "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "8d5fbc1a693fbfa9" }, { "scene_index": 7, "heading": "S#7. 유기견 보호원 / D", "beat_title": "사납게 짖는 로트와일러", "representative_moment": "로트와일러(C11)가 철창살 사이로 이빨을 드러내며 카메라를 향해 거칠게 짖고 있는 순간.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Low angle close-up at muzzle height, sickly green-gray fluorescent lighting", "t2i_prompt": "Photorealistic cinematic still. A low angle close-up at muzzle height. C11O00, a massive black Rottweiler, facing the camera, head pushing aggressively forward, thick neck and upper chest visible in the shadows behind. Thick, vertical metallic cage bars slice the immediate foreground, slightly rusted and scratched, separating the camera from the dog. The dog's open jaws, wet black nose, and exposed teeth dripping with saliva occupy the central composition, eyes fixed menacingly on the camera. Harsh overhead fluorescent lighting casts a cool, sterile green-gray wash over the black fur, creating sharp highlights on the teeth and wet saliva. [L09: dim, stained enclosure wall and a gritty wet concrete floor patch visible at the lower frame edge]. Focus on the terrifying intensity of the snarling head pushing against the metallic barrier.", "outfit_assignments": [ { "character_id": "C11", "outlook_id": "O00" } ], "source_facts": [ "S#7 action: '사납게 짖으며 날뛰는 로트와일러'", "Shot 1 description: '철장 안에서 이빨을 거칠게 드러내고 입을 벌린 거대한 검은색 로트와일러의 머리.'" ], "visual_inferences": [ "fixed_element kennel_cage_bars: Thick, vertical metallic cage bars divide the space, acting as a stark architectural element", "fixed_element sterile_fluorescent_lighting: harsh overhead fluorescent lights, casting a cool, sterile wash with subtle waxy, green-gray tones" ], "creative_decisions": [ "Low angle close-up directly at muzzle height to emphasize the intimidating size of the open jaws and exposed teeth", "Through-bars composition slicing the foreground to make the kennel feel dangerously narrow and immediate", "Frame-edge only background description to comply with close-framing background omission rules" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "dc589b8d9fb34517", "owned_hash": "24cb3d73858cd4a8", "camera_direction_hash": "0c33098dee57ef72", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "독립적으로 묘사 가능한 인트로성 스틸 컷", "_shot_index": 1, "visible_entities": [ "C11", "L09" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 7, "shot_index": 1 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Observer perspective, direct vision from just inside the kennel gate. A low angle CU sits at the 로트와일러’s muzzle height, tighter than the flow default so the open jaws, wet black nose, and exposed teeth occupy most of the center while vertical cage bars slice the foreground without exceeding the frame. The dog’s head pushes toward camera as a through-bars composition, making the kennel feel too narrow and immediate.", "lighting_mood": "Sickly shelter fluorescents flicker over black fur and saliva, leaving the kennel corners in green-gray shadow with harsh highlights on the teeth.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C11", "L09" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C11", "outlook_id": "O00" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_kennel_aisle_day_vaccination_busy", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "kennel_cage_bars", "element_type": "environment_state", "character_name": "", "description": "Thick, vertical metallic cage bars divide the space, acting as a stark architectural element that forms the animal enclosure. The dull metal surfaces create a rigid vertical pattern that visually separates the subjects.", "applies_to_shots": [ 1, 4 ], "source_facts": [ "Shot 1 description: 'vertical cage bars slice the foreground without exceeding the frame.'", "Shot 4 description: 'The kennel bars are now a soft peripheral pattern'" ], "visual_inferences": [ "Heavy-duty steel or aluminum bars with a slightly worn metallic finish, typical of enclosures holding large aggressive dogs." ], "creative_decisions": [ "Maintaining the vertical bars across both shots to establish the physical barrier of the kennel, while shifting their depth of field from foreground to background to match the camera's focus." ], "confidence": "high" }, { "element_id": "sterile_fluorescent_lighting", "element_type": "environment_state", "character_name": "", "description": "The interior is illuminated by harsh overhead fluorescent lights, casting a cool, sterile wash with subtle waxy, green-gray tones across the surfaces. This lighting creates an unglamorous and slightly unnerving clinical atmosphere uniformly throughout the facility.", "applies_to_shots": [ 1, 4, 8 ], "source_facts": [ "Shot 1: 'Sickly shelter fluorescents flicker... leaving the kennel corners in green-gray shadow.'", "Shot 4: 'Flat fluorescent light turns the fur and skin waxy, with a faint green cast.'", "Shot 8: 'Cool fluorescent wash reflects off water and steel... tender but still sterile.'" ], "visual_inferences": [ "Industrial fluorescent tube lighting mounted on the shelter ceiling, emitting a consistently harsh spectrum." ], "creative_decisions": [ "Using the consistent cool, green-tinted lighting to visually unify the different areas of the shelter (the kennel and the bathing station), reinforcing an unglamorous mood." ], "confidence": "high" } ], "previous_shot_refs": [], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C11O00", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "a38ca704fde925fc" }, { "scene_index": 7, "heading": "S#7. 유기견 보호원 / D", "beat_title": "주사 맞는 맹견과 수의사의 손", "representative_moment": "A middle-aged East Asian man's hand injecting a syringe into the flank of a subdued Rottweiler.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Extreme close-up with a slight Dutch angle, pushing in low beside the dog's rear flank, flat fluorescent lighting with a faint green-gray cast.", "t2i_prompt": "Photorealistic cinematic still. Extreme close-up with a slight Dutch angle, focused tightly on a middle-aged East Asian man's rough hand holding a medical syringe. The needle is inserted firmly into the pinched skin of a prone Rottweiler's rear flank in the central third of the frame. The dog's thick dark fur fills the immediate edges, providing scale. [L09: a damp, scuffed concrete floor visible just below the dog, and a soft out-of-focus peripheral pattern of thick vertical metallic cage bars behind]. Flat overhead fluorescent light reflects off the syringe barrel and turns the exposed skin waxy with a faint green-gray cast, creating an unnervingly sterile and clinical atmosphere.", "outfit_assignments": [], "source_facts": [ "Shot 4 description: 엎드린 개의 엉덩이에 주사기 바늘을 꽂은 50대 한국인 남성의 손 클로즈업.", "S7 action: 맹견의 엉덩이에 주사를 놓고 빠지는 50대 의사" ], "visual_inferences": [ "fixed_element sterile_fluorescent_lighting: harsh overhead fluorescent lights casting a cool, sterile wash with subtle waxy, green-gray tones.", "fixed_element kennel_cage_bars: thick, vertical metallic cage bars as a soft peripheral pattern.", "Unregistered 50s male doctor rendered as 'a middle-aged East Asian man's rough hand' following body-part focus and demographic descriptor rules." ], "creative_decisions": [ "Applied a slight Dutch angle to heighten the unnerving, clinical tension of the ECU.", "Used the L09 frame-edge block to define the immediate concrete floor and soft kennel bars, complying with the skipped_close_framing background policy without triggering whole-room reference matching." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "0bdf643613357618", "owned_hash": "24cb3d73858cd4a8", "camera_direction_hash": "a15e56e9b1456eba", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 7, "dependency_reason": "Same kennel environment and lighting as the prior shots in the scene, maintaining the continuous waxy green-gray fluorescent cast.", "_shot_index": 4, "visible_entities": [ "L09" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 7, "shot_index": 4 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Observer perspective, direct vision continuing the lateral medical-action flow. An ECU with a slight Dutch angle pushes in low beside the prone dog’s rear flank, isolating the syringe needle, pinched skin, and 50대 한국인 남성’s rough hand in the central third while the dog’s fur provides scale around the edges. The kennel bars are now a soft peripheral pattern, keeping the action grounded without pulling focus from the injection point.", "lighting_mood": "Flat fluorescent light turns the fur and skin waxy, with a faint green cast that makes the routine injection feel unnervingly surgical.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "L09" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C11", "outlook_id": "O00" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_kennel_aisle_day_vaccination_busy", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "kennel_cage_bars", "element_type": "environment_state", "character_name": "", "description": "Thick, vertical metallic cage bars divide the space, acting as a stark architectural element that forms the animal enclosure. The dull metal surfaces create a rigid vertical pattern that visually separates the subjects.", "applies_to_shots": [ 1, 4 ], "source_facts": [ "Shot 1 description: 'vertical cage bars slice the foreground without exceeding the frame.'", "Shot 4 description: 'The kennel bars are now a soft peripheral pattern'" ], "visual_inferences": [ "Heavy-duty steel or aluminum bars with a slightly worn metallic finish, typical of enclosures holding large aggressive dogs." ], "creative_decisions": [ "Maintaining the vertical bars across both shots to establish the physical barrier of the kennel, while shifting their depth of field from foreground to background to match the camera's focus." ], "confidence": "high" }, { "element_id": "sterile_fluorescent_lighting", "element_type": "environment_state", "character_name": "", "description": "The interior is illuminated by harsh overhead fluorescent lights, casting a cool, sterile wash with subtle waxy, green-gray tones across the surfaces. This lighting creates an unglamorous and slightly unnerving clinical atmosphere uniformly throughout the facility.", "applies_to_shots": [ 1, 4, 8 ], "source_facts": [ "Shot 1: 'Sickly shelter fluorescents flicker... leaving the kennel corners in green-gray shadow.'", "Shot 4: 'Flat fluorescent light turns the fur and skin waxy, with a faint green cast.'", "Shot 8: 'Cool fluorescent wash reflects off water and steel... tender but still sterile.'" ], "visual_inferences": [ "Industrial fluorescent tube lighting mounted on the shelter ceiling, emitting a consistently harsh spectrum." ], "creative_decisions": [ "Using the consistent cool, green-tinted lighting to visually unify the different areas of the shelter (the kennel and the bathing station), reinforcing an unglamorous mood." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 7, "shot_index": 1, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C11O00", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "f59192ed301e6bbb" }, { "scene_index": 7, "heading": "S#7. 유기견 보호원 / D", "beat_title": "수리영의 목욕 봉사", "representative_moment": "수리영(C01)이 작은 강아지의 등에 양손을 얹고 목욕을 시키는 정지 순간.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Observer perspective, slightly overhead medium shot at arm level", "t2i_prompt": "Photorealistic cinematic still. Observer perspective, slightly overhead medium shot at arm level. C01O02 in a navy windbreaker, a young East Asian woman, looking down, leaning slightly forward with her torso and forearms in frame. Her two hands are placed symmetrically from left and right on the foamy wet back of a small puppy in the center, her forearms framing the dog. Focus on the small puppy and the fresh white foam spread across its wet fur. The dog rests inside the wet stainless tub from the reference, water streaks visible below. The coiled nozzle of the sprayer from the reference points away from the camera at the rear edge. Harsh overhead fluorescent lights from the reference illuminate the interior, casting a cool, sterile wash with subtle waxy, green-gray tones across the surfaces, softened only by the white foam.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "S#7 action: ‘네’, 하고 달려가 털깍고, 연고 바르고, 목욕시키고 등", "shot description: 작은 강아지의 젖은 등에 비누 거품을 묻힌 채 양손을 얹은 수리영의 정지 구도.", "DP instructions: Observer perspective, slightly overhead MS detail at arm level... hands placed symmetrically... cool fluorescent wash softened by white foam." ], "visual_inferences": [ "fixed_element sterile_fluorescent_lighting: harsh overhead fluorescent lights casting a cool, sterile wash with waxy green-gray tones", "anchoring owned objects from background reference: the tub and the sprayer", "mid-action pause with symmetric hand placement conveys a shift to care while keeping the clinical atmosphere" ], "creative_decisions": [ "Framed from a slightly elevated angle to clearly show the hands framing the puppy and the foam, with the torso and arms of C01O02 providing the upper frame context without triggering an isolated body-part close-up", "Maintaining the background reference elements (tub, sprayer) to ensure continuity with the exact_background rule" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "63be1edf02ed486b", "owned_hash": "bc6a99e389261123", "camera_direction_hash": "0a3525d363e46e4e", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "tub", "violating_phrase": "The dog rests inside the wet stainless tub from the reference", "reason": "This explicitly reuses the existing tub from the reference rather than instructing a new tub to be created, so it is not a redraw violation." }, { "owned_object": "sprayer", "violating_phrase": "The coiled nozzle of the sprayer from the reference", "reason": "This explicitly refers to the sprayer already in the reference, which is allowed as an anchor reference." }, { "owned_object": "fluorescent light", "violating_phrase": "Harsh overhead fluorescent lights from the reference", "reason": "This explicitly refers to the existing fluorescent lights from the reference, not a new light being added." } ] } } ], "scene_type": "normal", "dependent_scene_index": 7, "dependency_reason": "S7_Shot4의 동일한 공간(exact_background) 참조 및 조명 톤 유지", "_shot_index": 8, "visible_entities": [ "C01", "L09" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 7, "shot_index": 8 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Observer perspective, direct vision at the end of the care-close-observation flow. A slightly overhead MS detail holds static at arm level above the bathing station, with 수리영’s two hands placed symmetrically from left and right on the small puppy’s foamy wet back in the center; her forearms frame the dog without letting any prop overwhelm the image. The composition settles into a quiet long-take implied pause, shifting the scene’s rhythm from threat to care.", "lighting_mood": "Cool fluorescent wash reflects off water and steel, softened by white foam so the horror palette briefly becomes tender but still sterile.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "L09" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C11", "outlook_id": "O00" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "bg_wash_grooming_area_day_bathing", "reference_usage": "exact_background", "owned_objects": [ "bottle", "counter", "door", "drain", "dryer blower", "drying rack", "faucet", "fluorescent light", "grooming table", "ramp", "shelf", "shelving", "sink", "sprayer", "stool", "towel", "tub", "ventilation fan", "washer dryer", "waste bin", "window" ], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "owned_objects lists all environment objects already drawn in the background reference — do not create new ones", "when focusing on an owned object, anchor it explicitly: 'the door from the reference'" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "sterile_fluorescent_lighting", "element_type": "environment_state", "character_name": "", "description": "The interior is illuminated by harsh overhead fluorescent lights, casting a cool, sterile wash with subtle waxy, green-gray tones across the surfaces. This lighting creates an unglamorous and slightly unnerving clinical atmosphere uniformly throughout the facility.", "applies_to_shots": [ 1, 4, 8 ], "source_facts": [ "Shot 1: 'Sickly shelter fluorescents flicker... leaving the kennel corners in green-gray shadow.'", "Shot 4: 'Flat fluorescent light turns the fur and skin waxy, with a faint green cast.'", "Shot 8: 'Cool fluorescent wash reflects off water and steel... tender but still sterile.'" ], "visual_inferences": [ "Industrial fluorescent tube lighting mounted on the shelter ceiling, emitting a consistently harsh spectrum." ], "creative_decisions": [ "Using the consistent cool, green-tinted lighting to visually unify the different areas of the shelter (the kennel and the bathing station), reinforcing an unglamorous mood." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 7, "shot_index": 4, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C11O00", "policy": "required" }, { "kind": "background", "id": "bg_wash_grooming_area_day_bathing", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "83d6604855d996cc", "status": "contract_violation" }, { "scene_index": 8, "heading": "S#8. 번화가·대형 마트 / 해질 무렵", "beat_title": "엇갈린 만남과 커지는 의문", "representative_moment": "직원의 말을 듣고 혼란에 빠져 눈을 동그랗게 뜬 수리영(C01)의 얼굴을 정면에서 포착한다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Eye-level close-up, sharp facial focus with blurred vertical color bands behind", "t2i_prompt": "Photorealistic cinematic still. An eye-level close-up of C01O02 in a navy windbreaker, a young East Asian woman, facing the camera with her chin slightly lifted. Her widened eyes are fixed on an off-camera middle-aged East Asian woman, her expression locked in sudden confusion. Her face and neck occupy the center frame, with her shoulder partially visible at the lower edge. The background is [L11: a tight frame of heavily stocked display shelves blurring into soft vertical color bands under slightly greenish-white overhead fluorescent ceiling fixtures]. Cold fluorescent light washes over her face, while a pervasive dusk-blue ambient glow faintly cools the shadows on her jawline and jacket collar, isolating her sharp focus against the blurred commercial environment.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "S#8 action: 수리영 realizes she missed her mom at the mart after hearing the coworker's statement.", "shot description: eye-level CU frames 수리영's face almost straight on, widened eyes, fluorescent-lit products blur into vertical color bands.", "Scene heading: 해질 무렵 (around dusk)" ], "visual_inferences": [ "fixed_element supermarket_interior_aisle: brightly illuminated by slightly greenish-white overhead fluorescent ceiling fixtures", "fixed_element dusk_evening_lighting: pervasive dusk-blue ambient glow", "The off-camera coworker's presence is implied purely through C01O02's eyeline" ], "creative_decisions": [ "Tight close-up framing with shoulder visible to avoid floating head artifact", "Shallow depth of field blurring the background shelves into soft vertical color bands to emphasize internal confusion", "Mixing the cold greenish-white fluorescent light on the face with dusk-blue ambient light in the shadows to create an uneasy mood" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "70d9a9b91d483369", "owned_hash": "d2085ab95968fafc", "camera_direction_hash": "0ff085efb9ca4552", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "No direct visual continuity required from a previous scene for this specific close-up moment.", "_shot_index": 3, "visible_entities": [ "C01", "P05", "L11", "L12" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 8, "shot_index": 3 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Observer perspective, direct vision inheriting the frontal close-observation stage. An eye-level CU frames 수리영’s face almost straight on, tightened so her widened eyes sit just above center while the 50대 한국인 여성 remains implied by her eyeline off-camera near the display shelf. Fluorescent-lit products blur into vertical color bands behind her, emphasizing that only her confusion has snapped into focus.", "lighting_mood": "Cold supermarket fluorescents flatten her face while dusk-blue spill from the entrance faintly cools the shadows, making the ordinary aisle feel uneasy.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "P05", "L11", "L12" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C06", "outlook_id": "O08" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "cb_sales_floor_dusk_normal", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "supermarket_interior_aisle", "element_type": "environment_state", "character_name": "", "description": "The interior of the large retail mart features rows of display shelves heavily stocked with various products. The space is brightly illuminated by slightly greenish-white overhead fluorescent ceiling fixtures, creating a cold, uniform commercial atmosphere.", "applies_to_shots": [ 3, 5 ], "source_facts": [ "Shot 3: Fluorescent-lit products blur into vertical color bands", "Shot 3: Cold supermarket fluorescents flatten her face", "Shot 5: Greenish-white fluorescent light mixes with faint sunset spill" ], "visual_inferences": [ "Standard dense product arrangement typical of a large Korean grocery store", "Continuous rows of fluorescent tubes running along the ceiling" ], "creative_decisions": [ "Emphasizing the harsh, sterile greenish-white lighting to contrast against the natural dusk light, subtly heightening the character's growing unease" ], "confidence": "high" }, { "element_id": "dusk_evening_lighting", "element_type": "environment_state", "character_name": "", "description": "The overall environmental lighting reflects early evening twilight, characterized by a pervasive dusk-blue ambient glow. This natural dim blue light fully wraps the exterior surroundings and faintly spills through the main entrance into the building's interior, intermingling with artificial lights.", "applies_to_shots": [ 3, 5, 6 ], "source_facts": [ "Scene heading: 해질 무렵 (around dusk)", "Shot 3: dusk-blue spill from the entrance faintly cools the shadows", "Shot 5: faint sunset spill from the aisle opening", "Shot 6: Dusk-blue ambient light wraps her profile" ], "visual_inferences": [ "The darkening sky outside casts a cool twilight shadow that penetrates the threshold of the automatic glass doors" ], "creative_decisions": [ "Using the dusk-blue tone as a consistent unifying atmospheric element across both interior observation and exterior profile shots" ], "confidence": "high" } ], "previous_shot_refs": [], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C06O08", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "031ee565382a0ce7" }, { "scene_index": 8, "heading": "번화가·대형 마트 / 해질 무렵", "beat_title": "엄마가 밖에서 누군가와 얘기했다는 소식을 듣고 의아해진 상태", "representative_moment": "남자 직원을 향해 고개를 살짝 기울인 수리영의 의아한 표정.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Eye-level medium close-up, over-the-shoulder composition with greenish-white fluorescent and dusk-blue contrast", "t2i_prompt": "Photorealistic cinematic still. Eye-level medium close-up. Focus on the questioning expression of C01O02 in a navy windbreaker, a young East Asian woman, occupying the left-center foreground at a three-quarter angle from the right, her upper body in frame. Her head is tilted slightly, eyes fixed on the man across the shared aisle space. In the right foreground edge, C06O08 in a red uniform, a middle-aged East Asian man, appears as a soft, out-of-focus shoulder slice at a three-quarter angle from the left, balancing stacked cardboard boxes mid-step, looking toward her. Greenish-white fluorescent light washes her cheek and jacket, mixing with a faint dusk-blue sunset spill falling on her shoulder. [L11: vertical rows of stocked shelves softly receding in the background, a bright aisle opening visible behind the right edge bathed in dim twilight blue.]", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C06", "outlook_id": "O08" } ], "source_facts": [ "Shot 5: 남자 직원을 향해 고개를 살짝 기울인 수리영의 의아한 표정", "Shot 5: Greenish-white fluorescent light mixes with faint sunset spill from the aisle opening", "Scene heading: 해질 무렵 (around dusk)" ], "visual_inferences": [ "fixed_element dusk_evening_lighting: The darkening sky outside casts a cool twilight shadow that penetrates the threshold", "fixed_element supermarket_interior_aisle: brightly illuminated by slightly greenish-white overhead fluorescent ceiling fixtures", "C06 balancing boxes mid-step as an action prop for a mart employee" ], "creative_decisions": [ "Eye-level MCU framing over the shoulder of C06 to emphasize C01's questioning expression", "Soft focus on C06 as a shoulder slice to adhere to close framing scale while grounding the spatial interaction across the shared aisle", "Greenish-white fluorescent light on her face to contrast with the dusk-blue ambient background, heightening tension" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "eefad2f32a08b795", "owned_hash": "d2085ab95968fafc", "camera_direction_hash": "da315d878b2e40bc", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 8, "dependency_reason": "Same retail mart interior location and dusk lighting continuity (previous shot 3)", "_shot_index": 5, "visible_entities": [ "C01", "C06", "P05", "L11", "L12" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 8, "shot_index": 5 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Observer perspective, direct vision from the aisle corner as the camera completes its pan. An eye-level MCU favors 수리영 from her front-left in a subtle over-the-shoulder/over-axis arrangement, her tilted head held in the left-center foreground while 남자 직원 stands off-axis to the right with boxes at chest height. The aisle opening stays visible behind them, giving the new information a direction of escape toward the outside.", "lighting_mood": "Greenish-white fluorescent light mixes with faint sunset spill from the aisle opening, increasing the sickly tension as her uncertainty sharpens.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "C06", "P05", "L11", "L12" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C06", "outlook_id": "O08" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "cb_sales_floor_dusk_normal", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "supermarket_interior_aisle", "element_type": "environment_state", "character_name": "", "description": "The interior of the large retail mart features rows of display shelves heavily stocked with various products. The space is brightly illuminated by slightly greenish-white overhead fluorescent ceiling fixtures, creating a cold, uniform commercial atmosphere.", "applies_to_shots": [ 3, 5 ], "source_facts": [ "Shot 3: Fluorescent-lit products blur into vertical color bands", "Shot 3: Cold supermarket fluorescents flatten her face", "Shot 5: Greenish-white fluorescent light mixes with faint sunset spill" ], "visual_inferences": [ "Standard dense product arrangement typical of a large Korean grocery store", "Continuous rows of fluorescent tubes running along the ceiling" ], "creative_decisions": [ "Emphasizing the harsh, sterile greenish-white lighting to contrast against the natural dusk light, subtly heightening the character's growing unease" ], "confidence": "high" }, { "element_id": "dusk_evening_lighting", "element_type": "environment_state", "character_name": "", "description": "The overall environmental lighting reflects early evening twilight, characterized by a pervasive dusk-blue ambient glow. This natural dim blue light fully wraps the exterior surroundings and faintly spills through the main entrance into the building's interior, intermingling with artificial lights.", "applies_to_shots": [ 3, 5, 6 ], "source_facts": [ "Scene heading: 해질 무렵 (around dusk)", "Shot 3: dusk-blue spill from the entrance faintly cools the shadows", "Shot 5: faint sunset spill from the aisle opening", "Shot 6: Dusk-blue ambient light wraps her profile" ], "visual_inferences": [ "The darkening sky outside casts a cool twilight shadow that penetrates the threshold of the automatic glass doors" ], "creative_decisions": [ "Using the dusk-blue tone as a consistent unifying atmospheric element across both interior observation and exterior profile shots" ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 8, "shot_index": 3, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C06O08", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "138f0aaa63ba67b2" }, { "scene_index": 8, "heading": "번화가·대형 마트 / 해질 무렵", "beat_title": "엄마와 엇갈려 연락을 시도하는 수리영", "representative_moment": "C01이 마트 밖에서 연락되지 않는 P05를 귀에 댄 채 불안하게 서 있는 순간.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Eye-level medium profile shot, dusk-blue ambient with flickering neon reflections", "t2i_prompt": "Photorealistic cinematic still. C01O02 in a navy windbreaker, a young East Asian woman, stands in right profile occupying the right third of the frame. She presses a smartphone to her ear, eyes fixed on the distance. Focus on her tense posture holding the phone. She stands on the damp sidewalk from the reference. In the middle background, the automatic glass doors from the reference sit softly behind her, reflecting blurred red-blue neon streaks and dusk traffic. A cool dusk-blue ambient twilight wraps her profile, while supermarket fluorescents from the interior and neon reflections add uneasy pink and cyan accents across the surrounding surfaces. Eye-level medium framing, upper body and head visible, capturing the isolating urban evening.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "Scene heading: 번화가·대형 마트 / 해질 무렵", "Action: 수리영 마트 밖을 나오면서 연락을 취해보지만 받지않는다.", "Shot description: 어스름이 깔린 마트 자동문 밖에서 휴대폰을 귀에 댄 수리영의 옆모습." ], "visual_inferences": [ "fixed_element dusk_evening_lighting: The natural dim blue light fully wraps the exterior surroundings and faintly spills through the main entrance", "Neon signs and traffic reflected on the glass doors create a slightly eerie, isolating urban texture." ], "creative_decisions": [ "Eye-level medium profile framing, placing her on the right third to emphasize her anticipation and the empty space.", "Dusk-blue ambient mixed with faint pink/cyan neon reflections for an uneasy mood." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "3ec5e9edff584662", "owned_hash": "f09a73301b99308c", "camera_direction_hash": "41d24a6df5c411b9", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "sidewalk", "violating_phrase": "She stands on the damp sidewalk from the reference.", "reason": "This refers to the existing sidewalk as a background anchor rather than instructing a new sidewalk to be created or placed." }, { "owned_object": "door", "violating_phrase": "the automatic glass doors from the reference sit softly behind her", "reason": "This explicitly references the doors already in the reference/background, not a new door to be drawn." }, { "owned_object": "glass panel", "violating_phrase": "reflecting blurred red-blue neon streaks", "reason": "This describes reflections on the existing glass surfaces rather than adding a new glass panel." } ] } } ], "scene_type": "normal", "dependent_scene_index": 8, "dependency_reason": "Using exact background reference from S8_Shot3 to maintain spatial continuity of the supermarket exterior and glass doors.", "_shot_index": 6, "visible_entities": [ "C01", "P05", "L11", "L12" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 8, "shot_index": 6 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Observer perspective, direct vision after the cut outside. An eye-level MS profile frames 수리영 four feet away on the sidewalk side, her side silhouette occupying the right third as she presses a phone to her ear; the automatic supermarket doors sit softly behind her in the middle background. The phone remains small and realistic near her cheek, while reflected neon and dusk traffic flicker across the glass without taking over the frame.", "lighting_mood": "Dusk-blue ambient light wraps her profile while supermarket fluorescents and neon reflections add uneasy pink and cyan accents behind her.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "P05", "L11", "L12" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C06", "outlook_id": "O08" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "cb_exterior_entrance_dusk_phone", "reference_usage": "exact_background", "owned_objects": [ "CCTV camera", "alcove", "ceiling light", "curb", "door", "door sensor", "floor mat", "glass panel", "sidewalk", "storefront", "street", "wall" ], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "owned_objects lists all environment objects already drawn in the background reference — do not create new ones", "when focusing on an owned object, anchor it explicitly: 'the door from the reference'" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "dusk_evening_lighting", "element_type": "environment_state", "character_name": "", "description": "The overall environmental lighting reflects early evening twilight, characterized by a pervasive dusk-blue ambient glow. This natural dim blue light fully wraps the exterior surroundings and faintly spills through the main entrance into the building's interior, intermingling with artificial lights.", "applies_to_shots": [ 3, 5, 6 ], "source_facts": [ "Scene heading: 해질 무렵 (around dusk)", "Shot 3: dusk-blue spill from the entrance faintly cools the shadows", "Shot 5: faint sunset spill from the aisle opening", "Shot 6: Dusk-blue ambient light wraps her profile" ], "visual_inferences": [ "The darkening sky outside casts a cool twilight shadow that penetrates the threshold of the automatic glass doors" ], "creative_decisions": [ "Using the dusk-blue tone as a consistent unifying atmospheric element across both interior observation and exterior profile shots" ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 8, "shot_index": 3, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C06O08", "policy": "required" }, { "kind": "background", "id": "cb_exterior_entrance_dusk_phone", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "ba33c1881f482fc1", "status": "contract_violation" }, { "scene_index": 9, "heading": "S#9. 해안가 / N", "beat_title": "해안도로를 달리는 수리영", "representative_moment": "차가운 달빛을 등지고 가파른 경사로를 오르는 C01의 자전거 실루엣.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Low angle observer perspective, underexposed blue-black moon backlight", "t2i_prompt": "Photorealistic cinematic still. C01O02 in a navy windbreaker, a young East Asian woman, in left profile, seated on the saddle of a bicycle mid-pedal climbing an uphill asphalt road, torso pitched forward at the strongest stroke, one knee bent sharply, eyes focused up the steep incline. Her hair flows back in the direction opposite to motion, the bicycle frame and wheels clearly visible beneath her. She appears as a backlit figure against the bright cold moon, rim light tracing her shoulder and profile, body volume preserved against the dim ambient fill, face in deep shadow but jaw contour identifiable. Low angle observer perspective, the dark damp road rising diagonally from the lower right to the upper left, her lower legs and the bicycle tires occupying the lower frame. Dense blue-gray coastal haze blooming around the edges, obscuring the seaward horizon, the underexposed moonlight casting the scene into a stark blue-black mood. Focus on the steep diagonal incline of the asphalt road and the bicycle's upward trajectory.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "S#9 action: 수리영 rushing home on a bicycle along a coastal road.", "shot description: Silhouette of 수리영 and her bicycle climbing a steep road against a cold moon, mid-pedal." ], "visual_inferences": [ "low angle capturing the upward diagonal force of the road with bicycle tires occupying the lower frame", "rim light preserving body volume and jaw contour to avoid flat featureless cutout", "mid-pedal freeze with hair flowing backward opposite to the direction of motion" ], "creative_decisions": [ "underexposed blue-black monochrome palette to emphasize the cold night mood", "dense coastal haze used as negative space to isolate the silhouette" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "ee9501a27095fab2", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "99e9552e6d3c462b", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 6, "dependency_reason": "동일한 해안도로 공간의 연장선이나, background mode not_applicable 이므로 이 샷의 순간과 분위기를 새롭게 묘사함", "_shot_index": 2, "visible_entities": [ "C01", "L08" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 9, "shot_index": 2 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Observer perspective, direct vision from the seaward shoulder. A low angle WS tracks three to five meters behind and beside 수리영, exposing for the cold moon so she and the bicycle become a hard black silhouette climbing the steep road across the center-left of frame. Her torso is pitched forward at the strongest pedal stroke, one knee bent sharply, with the uphill asphalt cutting diagonally toward the moonlit haze.", "lighting_mood": "Underexposed moon backlight carves everything into blue-black silhouette, with coastal haze blooming around the edges like an omen.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "L08" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "not_applicable", "bg_id": null, "reference_usage": "none", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "no chain_bg is bound to this shot — describe the location freely without referencing a background image", "do not use 'from the reference' / 'the existing room' / 'preserving the room' style phrasing — there is no reference image to refer to" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [ { "scene_index": 6, "shot_index": 2, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "287fe4aed51c0d5f" }, { "scene_index": 10, "heading": "S#10. 옥탑방 / N", "beat_title": "침입의 시작", "representative_moment": "낯선 남자의 손이 전단지를 철문 틈으로 꾹 밀어넣어 잠금을 풀고 있다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Eye-level extreme close-up detail shot, dirty blue-black shadow with sharp rust accent", "t2i_prompt": "Photorealistic cinematic still. Extreme close-up. Focus on the rough hand of a middle-aged East Asian man extending from a black trench coat sleeve. The fingers tightly grip a creased paper flyer, wedging its edge into the narrow vertical seam of a dark, scratched iron door facing the camera. The paper flyer is angled like a thin blade entering the latch line, compressed into the center of the frame. A weak rooftop practical light and pale moon spill catch the knuckles and the paper surface. The rest of the iron door drops into blackened negative space with deep, dirty blue-black shadows and a sharp crimson-brown rust accent along the edge.", "outfit_assignments": [], "source_facts": [ "S#10 action: '전단지를 뽑아서 손가락으로 문틈에 꾹 밀어 넣는다.'", "shot description: Close-up of a rough man's hand pushing a paper flyer into the tightly closed iron door's seam." ], "visual_inferences": [ "Body part focus rule applied: C10 replaced with 'rough hand of a middle-aged East Asian man' due to hand close-up constraint.", "Black trench coat sleeve visible at the frame edge to indicate C10's assigned outfit.", "Close framing omits the background context entirely, focusing only on the door seam and the intrusion tool." ], "creative_decisions": [ "Extreme close-up framing isolates the mechanical tension of the intrusion.", "Deep blue-black shadows and crimson-brown rust accents emphasize the gritty, dangerous atmosphere of the horror genre." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "a58209077dfc7dd6", "owned_hash": "17c170cf5d0badbe", "camera_direction_hash": "216bdec6405cd591", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 4, "dependency_reason": "S4_Shot1 is listed as a prior location reference, but this shot uses skipped_close_framing, so the background context is intentionally omitted from the generated composition.", "_shot_index": 3, "visible_entities": [ "C10", "P04", "P07", "L04" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 10, "shot_index": 3 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Observer perspective, direct vision at the iron door’s seam. An eye-level ECU dollies inches toward the narrow gap, with 거친 남자’s hand, the paper flyer, and the dark door seam compressed into the center while the rest of the iron door drops into blackened negative space. The flyer is angled like a thin blade entering the latch line, but it stays proportionate and framed by the scratched metal surface for scale.", "lighting_mood": "A weak rooftop practical and moon spill skim the hand and paper, leaving the iron door in dirty blue-black shadow with a sharp crimson-brown rust accent.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C10", "P04", "P07", "L04" ], "allowed_outlook_pairs": [ { "character_id": "C05", "outlook_id": "O06" }, { "character_id": "C10", "outlook_id": "O12" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_rooftop_night_intrusion_door_shadow", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [ { "scene_index": 4, "shot_index": 1, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C05O06", "policy": "required" }, { "kind": "character_outlook", "id": "C10O12", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "1369eef017322468" }, { "scene_index": 10, "heading": "S#10. 옥탑방 / N", "beat_title": "어둠 속의 습격", "representative_moment": "A young man bites the older woman's shoulder in the dark rooftop room.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Handheld close-up with a slight Dutch angle, focus on the shoulder contact point, cold blue-black shadows", "t2i_prompt": "Photorealistic cinematic still. Focus on the tense shoulder of a middle-aged East Asian woman in a rumpled shirt forming the foreground plane in left profile, her shoulder twisting away, eyes staring into the void. Entering from the dark left edge, the lower face of a young East Asian man in a dark jacket is captured at a three-quarter angle from the right, leaning into a violent bite, his eyes fixed on the woman. A weak strip of cold window spill reflects on the man's teeth and the woman's taut skin at the center-right of the frame. The remaining areas are swallowed by heavy cold blue-black shadows. [: nearly black, rough wallpaper texture barely visible at the frame edge]. Handheld close-up with a slight Dutch angle.", "outfit_assignments": [], "source_facts": [ "S#10 action: '민숙을 덮쳐 어깨를 물어뜯는 듯한 모습'", "Shot 7 description: 'A handheld CU with a slight Dutch angle pauses almost still at shoulder height beside the woman, angled across her shoulder toward the man’s mouth as it meets the skin in the center-right of frame.'" ], "visual_inferences": [ "fixed_element dark_rooftop_room_interior: 'heavy, cold blue-black shadows'", "The woman's twisted posture and tense skin convey the sudden violence of the attack" ], "creative_decisions": [ "Close framing on the shoulder contact point to emphasize the terrifying bite without revealing full faces, adhering to body-part focus rules", "Used common nouns with demographic descriptors for the characters as requested to omit specific IDs", "Replaced L01 with L02 for the frame-edge background to comply with ID removal requests while maintaining the skipped close framing environment", "Cold window spill lighting applied only to the surface of the teeth and skin to prevent rendering glitches" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "1e3ef2510be34704", "owned_hash": "17c170cf5d0badbe", "camera_direction_hash": "67ea0d30f1be35f7", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 4, "dependency_reason": "Maintain the established cold blue-black color palette from previous interior shots.", "_shot_index": 7, "visible_entities": [ "P04", "P07", "L04" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 10, "shot_index": 7 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Observer perspective, direct vision inside the cramped room. A handheld CU with a slight Dutch angle pauses almost still at shoulder height beside 한국인 여자, angled across her shoulder toward 한국인 남자’s mouth as it meets the skin in the center-right of frame. The woman’s shoulder and neck form the foreground plane, while the man’s lower face enters from the dark left edge, making the contact point the only sharp, terrifying detail before the scream.", "lighting_mood": "Near-dark interior lighting uses a narrow strip of cold window spill to catch skin and teeth, with the rest swallowed by blue-gray shadow and a sickly, violent stillness.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "P04", "P07", "L04" ], "allowed_outlook_pairs": [ { "character_id": "C05", "outlook_id": "O06" }, { "character_id": "C10", "outlook_id": "O12" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_rooftop_night_intrusion_door_shadow", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "dark_rooftop_room_interior", "element_type": "environment_state", "character_name": "", "description": "The cramped interior of the room is engulfed in heavy, cold blue-black shadows. A weak, sickly strip of light spills from an unseen window, cutting through the intense darkness and faintly illuminating the soft, dingy texture of a dirty wall in the background.", "applies_to_shots": [ 7, 8 ], "source_facts": [ "Shot 7 lighting: 'Near-dark interior lighting uses a narrow strip of cold window spill to catch skin and teeth, with the rest swallowed by blue-gray shadow'", "Shot 8 lighting: 'Cold blue-black room darkness is cut by a weak sickly spill from outside'", "Shot 8 camera: 'soft, dirty wall blur'" ], "visual_inferences": [ "The weak light source mimics nighttime streetlamps or faint moonlight, emphasizing the neglected and isolated nature of the rooftop room." ], "creative_decisions": [ "Maintaining a consistent cold blue-black color palette across the interior shots to establish a claustrophobic, violent, and terrifying atmosphere." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 4, "shot_index": 1, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C05O06", "policy": "required" }, { "kind": "character_outlook", "id": "C10O12", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "0da72a75f8d8c2be" }, { "scene_index": 10, "heading": "S#10. 옥탑방 / N", "beat_title": "처참한 비명", "representative_moment": "어둠 속에서 고통에 차 눈을 질끈 감고 비명을 지르는 C10의 얼굴 클로즈업.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Extreme close-up from slightly below eyeline looking up, with a cold blue-black cinematic color grade", "t2i_prompt": "Photorealistic cinematic still. Head-and-shoulder composition. in a dark casual top, a middle-aged East Asian woman, facing the camera, head thrown back, with eyes closed in sheer agony. The camera is positioned slightly below her eyeline looking up, pushing her face toward the center frame with her tense neck and shoulder slice visible at the bottom edge. Sharp focus on her stretched mouth frozen in a scream and deeply clenched eyelids. Sweat-darkened skin texture is highly detailed. The attacker is completely excluded. [: soft, dirty wall blur, engulfed in heavy, cold blue-black shadows directly behind her]. A weak, sickly strip of cold light falling across her skin, contrasting with a raw crimson warmth reflecting on her lower jaw. The deep shadows swallow the edges of the frame, emphasizing the claustrophobic space.", "outfit_assignments": [], "source_facts": [ "Shot 8 action: '민숙을 덮쳐 어깨를 물어뜯는 듯한 모습 ‘아악-’'", "Shot 8 description: '눈을 질끈 감고 입을 크게 벌린 한국인 여성의 고통스러운 얼굴 클로즈업'", "Shot 8 camera: 'From inside the dark rooftop room at face height and slightly below the woman's eyeline... holding only her clenched eyelids, stretched mouth... attacker is completely excluded'" ], "visual_inferences": [ "fixed_element 'dark_rooftop_room_interior' applied: cold blue-black room darkness and a weak sickly strip of light falling across the skin", "The head-and-shoulder composition preserves the upper body partially in frame to prevent disembodiment while satisfying the extreme close-up requirement" ], "creative_decisions": [ "Positioning the camera slightly below eyeline looking up to heighten the upward pressure and tension of the scream", "Contrasting the cold blue-black ambient shadows with a raw crimson warmth reflecting on her jawline to highlight the violent atmosphere without showing the violence itself" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "fe4322ff11795844", "owned_hash": "17c170cf5d0badbe", "camera_direction_hash": "8f1fd5477665fd93", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 10, "dependency_reason": "S10_Shot7에서 설정된 옥탑방 내부의 어둡고 차가운 조명 톤과 좁고 폐쇄적인 분위기를 연속해서 유지", "_shot_index": 8, "visible_entities": [ "P04", "P07", "L04" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 10, "shot_index": 8 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From inside the dark rooftop room at face height and slightly below the woman's eyeline, the pan dies into an extreme close-up (ECU), eye-level with a slight upward pressure, holding only her clenched eyelids, stretched mouth, and sweat-dark skin texture. This is direct observer vision, not the attacker's POV; the attacker is completely excluded, and her face occupies most of the frame against a soft, dirty wall blur.", "lighting_mood": "Cold blue-black room darkness is cut by a weak sickly spill from outside, with a raw crimson warmth implied in the skin and mouth.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "P04", "P07", "L04" ], "allowed_outlook_pairs": [ { "character_id": "C05", "outlook_id": "O06" }, { "character_id": "C10", "outlook_id": "O12" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_rooftop_night_intrusion_door_shadow", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "dark_rooftop_room_interior", "element_type": "environment_state", "character_name": "", "description": "The cramped interior of the room is engulfed in heavy, cold blue-black shadows. A weak, sickly strip of light spills from an unseen window, cutting through the intense darkness and faintly illuminating the soft, dingy texture of a dirty wall in the background.", "applies_to_shots": [ 7, 8 ], "source_facts": [ "Shot 7 lighting: 'Near-dark interior lighting uses a narrow strip of cold window spill to catch skin and teeth, with the rest swallowed by blue-gray shadow'", "Shot 8 lighting: 'Cold blue-black room darkness is cut by a weak sickly spill from outside'", "Shot 8 camera: 'soft, dirty wall blur'" ], "visual_inferences": [ "The weak light source mimics nighttime streetlamps or faint moonlight, emphasizing the neglected and isolated nature of the rooftop room." ], "creative_decisions": [ "Maintaining a consistent cold blue-black color palette across the interior shots to establish a claustrophobic, violent, and terrifying atmosphere." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 10, "shot_index": 7, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C05O06", "policy": "required" }, { "kind": "character_outlook", "id": "C10O12", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "a0583fa8d3fec88d" }, { "scene_index": 11, "heading": "S#11. 옥탑 방 밖 / N", "beat_title": "불길한 조짐과 직면", "representative_moment": "C01 stands rigid, gazing intensely off-camera at the unseen doorway.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "medium close-up at a slight low angle, muted coastal night blues with harsh sickly fluorescent highlights", "t2i_prompt": "Photorealistic cinematic still. C01O02 in a navy windbreaker, a young East Asian woman, positioned at three-quarter angle from the left, her head and rigid shoulders occupying the center frame. Her eyes are fixed off-camera past the lens line. Focus on her tense expression in a slight low angle composition. A faint, sickly fluorescent light spills from an unseen doorway located just outside the frame, pooling on the side of her face and shoulder to cast a harsh highlight against the gloomy ambient shadows. The background is [L04: cold gray-blue coastal night darkness, a dark, partially fogged railing visible at the frame edge behind her].", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "S#11 action: 옥상에 올라와 보니 반쯤 열린 현관 철문.", "Shot 4 description: 옥상 바닥에 두 발을 딛고 선 채 앞을 뚫어지게 주시하는 수리영의 상체.", "DP instruction: MCU at a slight low angle, shoulders rigid, gaze cuts past the lens line." ], "visual_inferences": [ "fixed_element eerie_doorway_lighting: cold gray-blue coastal night darkness, faint sickly fluorescent highlight.", "Due to MCU framing and close_framing_skips_background_ref, the background is reduced to immediate frame-edge darkness and hazy railing.", "Feet are excluded geometrically from the MCU framing, focusing on upper body rigidity." ], "creative_decisions": [ "Slight low angle to emphasize tension and shock as she looks at the unseen door.", "Harsh contrast between cool night ambient and sickly pale spill on her skin." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "4eb999a4e1985434", "owned_hash": "98a58c7b3b700047", "camera_direction_hash": "48afc9eef4d16e7c", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "No dependent scene specified", "_shot_index": 4, "visible_entities": [ "C01", "L04", "L05", "L06" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 11, "shot_index": 4 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "On the rooftop floor about two meters in front of 수리영, the camera holds a medium close-up (MCU) at a slight low angle, derived from the chest-height frontal axis of the flow. 수리영 is centered but turned just off-axis, her shoulders rigid in the lower half while her gaze cuts past the lens line toward the half-open iron door behind camera.", "lighting_mood": "Muted coastal night blues flatten the rooftop while a faint fluorescent spill from the doorway gives her skin a sickly pallor.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "L04", "L05", "L06" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_approach_night_open_door", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "eerie_doorway_lighting", "element_type": "environment_state", "character_name": "", "description": "The surrounding environment is immersed in a cold, muted gray-blue coastal night darkness. A faint, sickly fluorescent light spills from an unseen, half-open doorway located just outside the frame, casting a distinct, harsh highlight against the gloomy ambient shadows.", "applies_to_shots": [ 4, 6 ], "source_facts": [ "Shot 4 lighting: Muted coastal night blues flatten the rooftop while a faint fluorescent spill from the doorway gives her skin a sickly pallor.", "Shot 6 lighting: A cold gray-blue top wash and a thin, dirty doorway highlight catch in her pupils." ], "visual_inferences": [ "The off-camera fluorescent light serves as the primary practical highlight source, consistently illuminating subjects facing the unseen half-open door." ], "creative_decisions": [ "Maintaining a stark contrast between the cool night ambient wash and the unnerving, pale doorway spill to unify the visual tone across different focal lengths." ], "confidence": "high" } ], "previous_shot_refs": [], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "3ac7b697e2edd115" }, { "scene_index": 11, "heading": "S#11. 옥탑 방 밖 / N", "beat_title": "반쯤 열린 철문을 발견하고 경악하는 수리영의 눈빛", "representative_moment": "Extreme close-up of C01's widened eyes, reflecting the eerie light from the half-open door.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Extreme close-up (ECU) on eyes, eye-level observer POV, cold gray-blue ambient with pale fluorescent highlights", "t2i_prompt": "Photorealistic cinematic still. Extreme close-up tight on the widened, terrified eyes of a young East Asian woman facing the camera. Her brow and the bridge of her nose form the tight borders of the frame. A cold, muted gray-blue ambient night wash covers her skin, while a faint, sickly pale fluorescent light from an unseen source falls across the side of her face. A thin highlight catches in her dilated pupils, reflecting a tiny glint of an open iron door. The extreme shallow depth of field blurs the immediate edges of her face into dark shadow, focusing entirely on the wet, tense gaze.", "outfit_assignments": [], "source_facts": [ "S#11 action: 수리영 spots the half-open iron door and realizes something is wrong.", "Shot 6 description: ECU on eyes, eye-level, widened eyes dominating the frame while the bridge of her nose and brow form tight borders." ], "visual_inferences": [ "fixed_element eerie_doorway_lighting: cold gray-blue night ambient and off-camera faint fluorescent doorway spill", "Clothing and background are completely omitted due to the extreme close-up framing on the eyes.", "The open iron door is only visible as a micro-reflection in the pupils, keeping the shot strictly subjective." ], "creative_decisions": [ "Applied body_part_focus_rule by replacing the C01O02 composite ID with 'young East Asian woman' to prevent face-reference scaling issues in extreme close-ups.", "Strict adherence to skipped_close_framing by isolating the face and omitting any spatial background or room references." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "a04167b0959b6323", "owned_hash": "98a58c7b3b700047", "camera_direction_hash": "8c20f6748800d152", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 11, "dependency_reason": "Same location and consecutive moment from Shot 4. Background is skipped due to ECU framing, but the established eerie_doorway_lighting carries over.", "_shot_index": 6, "visible_entities": [ "C01", "L04", "L05", "L06" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 11, "shot_index": 6 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Continuing along the same frontal axis, the dolly-in ends in an extreme close-up (ECU), eye-level, with both of 수리영's widened eyes dominating the frame while the bridge of her nose and brow form tight borders. The shot remains direct observer vision; her pupils are fixed beyond the camera toward the open iron door, and the rooftop background disappears into shallow darkness.", "lighting_mood": "A cold gray-blue top wash and a thin, dirty doorway highlight catch in her pupils, making the eyes feel wet and terrified.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "L04", "L05", "L06" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_approach_night_open_door", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "eerie_doorway_lighting", "element_type": "environment_state", "character_name": "", "description": "The surrounding environment is immersed in a cold, muted gray-blue coastal night darkness. A faint, sickly fluorescent light spills from an unseen, half-open doorway located just outside the frame, casting a distinct, harsh highlight against the gloomy ambient shadows.", "applies_to_shots": [ 4, 6 ], "source_facts": [ "Shot 4 lighting: Muted coastal night blues flatten the rooftop while a faint fluorescent spill from the doorway gives her skin a sickly pallor.", "Shot 6 lighting: A cold gray-blue top wash and a thin, dirty doorway highlight catch in her pupils." ], "visual_inferences": [ "The off-camera fluorescent light serves as the primary practical highlight source, consistently illuminating subjects facing the unseen half-open door." ], "creative_decisions": [ "Maintaining a stark contrast between the cool night ambient wash and the unnerving, pale doorway spill to unify the visual tone across different focal lengths." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 11, "shot_index": 4, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "6daeb39cb0240a6f" }, { "scene_index": 12, "heading": "S#12. 옥탑방 안 / N", "beat_title": "방문 틈새로 보이는 핏자국 발견", "representative_moment": "A close view through the partially open door reveals a terrifying trail of fresh blood on the floor.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "knee height observer POV, Dutch angle, narrow door gap through-object framing, cold blue shadow", "t2i_prompt": "Photorealistic cinematic still. Focus on the surface of the floor violently covered in fresh, vivid red bloodstains and smeared pools of blood. Framed from knee height at a tilted Dutch angle through a narrow door gap, the dark dirty edge of the door and the jamb slice the immediate foreground into thick vertical bands. Beyond the threshold in the center-right, darker, heavier streaks of blood form a disturbing trail across the muted flooring. The blood maintains a wet, glistening appearance. Sickly interior spill and cold blue shadows make the crimson look unnaturally vivid against the pale surface. [L07: Narrow gap through a partially open wooden door, muted floor surface bathed in cold shadows and sickly dim light]", "outfit_assignments": [], "source_facts": [ "Scene text: '여기저기 핏자국. 바닥에 흥건한 혈흔. 피묻은 발자국은 커튼으로 이어져 있다.'", "Shot 7 description: 방문 틈새로 보이는 방바닥 위 선명하고 붉은 핏자국 클로즈업." ], "visual_inferences": [ "fixed_element bloody_bedroom_floor: fresh, wet blood creating glossy reflections on the muted floor", "The narrow gap implies the door is only slightly open, restricting the view to the floor trail." ], "creative_decisions": [ "Dutch angle and knee-height framing amplify the horror and disorientation of the discovery.", "Cold blue shadows heavily contrasted with unnaturally vivid red to emphasize the disturbing, eerie atmosphere.", "Narrow through-door gap composition used to anchor the observer's restricted, tense point of view." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "eaf4ea79fabaec57", "owned_hash": "50e0a1bb6b2bd65b", "camera_direction_hash": "7f93a242364de21c", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 5, "dependency_reason": "Same rooftop room interior. Background reference from Scene 5 Shot 14 is policy-skipped due to close framing, but the spatial continuity of the muted floor surface is preserved.", "_shot_index": 7, "visible_entities": [ "P05", "P08", "L07" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 12, "shot_index": 7 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At knee height outside 수리영's bedroom doorway, the camera compresses into a close-up (CU), Dutch angle, using the narrow door gap as through-object framing. The red blood marks sit in the center-right of the floor beyond the threshold, while the dirty door edge and jamb slice the foreground into dark vertical bands; the implied dolly direction is carried by the trail leading deeper inside.", "lighting_mood": "Sickly interior spill and cold blue shadow make the fresh red blood look unnaturally vivid against the muted floor.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "P05", "P08", "L07" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C05", "outlook_id": "O06" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_daughter_bedroom_night_blood_curtain_body_red_circle", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "bloody_bedroom_floor", "element_type": "environment_state", "character_name": "", "description": "The surface of the room's floor is violently covered in fresh, vivid red bloodstains and smeared pools of blood. Darker, heavier streaks of blood form a disturbing trail across the muted flooring, maintaining a wet, glistening appearance under the dim interior light.", "applies_to_shots": [ 7, 10 ], "source_facts": [ "Scene text: '여기저기 핏자국. 바닥에 흥건한 혈흔. 피묻은 발자국은 커튼으로 이어져 있다.'", "Shot 7: 'The red blood marks sit in the center-right of the floor beyond the threshold'", "Shot 10: 'smeared floor remain blurred at the frame edges', 'crimson floor reflections flickering'" ], "visual_inferences": [ "The floor is made of a standard, muted material (like linoleum or wood) typical of a Korean rooftop room, providing a stark contrast to the red blood.", "The blood is fresh and wet, creating glossy reflections that catch the ambient room light." ], "creative_decisions": [ "Using the vibrant crimson color and smeared texture of the blood to anchor the spatial continuity and horror atmosphere across different camera angles in the room." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 5, "shot_index": 14, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C05O06", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "55fc74ea9140edde" }, { "scene_index": 12, "heading": "S#12. 옥탑방 안 / N", "beat_title": "수리영의 경악", "representative_moment": "C01 discovers the gruesome scene, her face frozen in absolute horror as she recoils.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Handheld close-up (CU), Dutch angle, frontal observer POV. Cold window haze and yellow-green room spill with crimson reflections.", "t2i_prompt": "Photorealistic cinematic still. Close-up on the horrified expression of C01O02 in a navy windbreaker, a young East Asian woman. Dutch angle, head-and-shoulder composition. She is facing the camera, recoiling mid-breath, her mouth open and eyes blown wide in sheer shock, eyes fixed on an unseen horror just out of frame. Cold haze and weak yellow-green ambient light wash across her face, while wet crimson light reflects faintly on her lower jaw and the edge of her collar. [L07: heavily out-of-focus background with a parted curtain hanging unevenly behind her shoulder, and glistening red smeared reflections on the dark surface barely visible along the bottom frame edge].", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "Scene text: '수리영 방... 헉!- 여기저기 핏자국. 바닥에 흥건한 혈흔... 앉은 채로 고개를 떨군 채 죽어있는 엄마- 민숙!!!!'", "Shot description: 수리영's face fills the center, eyes blown wide and mouth open... recoiling mid-breath" ], "visual_inferences": [ "fixed_element bloody_bedroom_floor: fresh, vivid red bloodstains creating glossy reflections at the lower frame edge", "Head-and-shoulder framing prevents face-floating artifacts while keeping the primary focus entirely on her shocked reaction." ], "creative_decisions": [ "Dutch angle applied to convey psychological unbalance and intense horror.", "Crimson reflections from the floor catching on her jawline to physically tie her shock to the bloody scene below without needing a wide shot." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "d477f61c91c2dae6", "owned_hash": "9f4861cca11ac4b0", "camera_direction_hash": "ccc1472e5f0edd23", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 12, "dependency_reason": "Maintains spatial and atmospheric continuity with the bloody room elements established in S12 Shot 7.", "_shot_index": 10, "visible_entities": [ "C01", "P05", "P08", "L07" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 12, "shot_index": 10 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Inside 수리영's bedroom near the curtained window, the handheld camera pushes into a close-up (CU), Dutch angle, frontal enough to trap her shock but trembling with her breath. 수리영's face fills the center, eyes blown wide and mouth open, while the parted curtain and smeared floor remain blurred at the frame edges as evidence of what she has just uncovered.", "lighting_mood": "The handheld close-up is lit by cold window haze and weak yellow-green room spill, with crimson floor reflections flickering at the bottom edge.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "P05", "P08", "L07" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C05", "outlook_id": "O06" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_living_kitchen_night_aftermath_tv_cups", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "bloody_bedroom_floor", "element_type": "environment_state", "character_name": "", "description": "The surface of the room's floor is violently covered in fresh, vivid red bloodstains and smeared pools of blood. Darker, heavier streaks of blood form a disturbing trail across the muted flooring, maintaining a wet, glistening appearance under the dim interior light.", "applies_to_shots": [ 7, 10 ], "source_facts": [ "Scene text: '여기저기 핏자국. 바닥에 흥건한 혈흔. 피묻은 발자국은 커튼으로 이어져 있다.'", "Shot 7: 'The red blood marks sit in the center-right of the floor beyond the threshold'", "Shot 10: 'smeared floor remain blurred at the frame edges', 'crimson floor reflections flickering'" ], "visual_inferences": [ "The floor is made of a standard, muted material (like linoleum or wood) typical of a Korean rooftop room, providing a stark contrast to the red blood.", "The blood is fresh and wet, creating glossy reflections that catch the ambient room light." ], "creative_decisions": [ "Using the vibrant crimson color and smeared texture of the blood to anchor the spatial continuity and horror atmosphere across different camera angles in the room." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 12, "shot_index": 7, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C05O06", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "63d718ac2eace0ff" }, { "scene_index": 12, "heading": "S#12. 옥탑방 안 / N", "beat_title": "벽에 나타난 기이한 붉은 원", "representative_moment": "텅 빈 흰색 벽지 한가운데 붉은 원형 표식이 젖은 피처럼 선명하게 새겨져 있다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Eye-level wide shot perfectly centered on the wall, flat cold interior light", "t2i_prompt": "Photorealistic cinematic still. Opposite the bed at eye height, the camera locks into a wide composition, perfectly centered on the blank white wall from the reference. Focus on the large, vivid red circular symbol from the reference sitting in the middle of the frame, looking like fresh, wet paint. The bed from the reference forms a low horizontal base in the lower frame. No characters are visible. Flat cold interior light drains the wall to bone-white, making the deeply saturated red circle appear occult and projected into reality, while the dim, symmetrical bedroom corners maintain rigid room geometry.", "outfit_assignments": [], "source_facts": [ "shot description: A still composition of a large, red, wet-looking circular symbol drawn on the blank white wallpaper above the bed. No characters visible.", "S#12 action: '침대 위 벽에 붉은 원이 생겨난다'" ], "visual_inferences": [ "fixed_element exact_background: S12_Shot7", "symmetrical bedroom corners", "flat cold interior lighting" ], "creative_decisions": [ "centered wide framing to emphasize the stark contrast and occult nature of the red symbol", "bone-white wall against saturated wet red paint effect" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "2b425b4b751ce184", "owned_hash": "50e0a1bb6b2bd65b", "camera_direction_hash": "0c2ec342c0f79d05", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "circle", "violating_phrase": "Focus on the large, vivid red circular symbol from the reference sitting in the middle of the frame", "reason": "This describes a new circular object/symbol to be rendered rather than merely referencing an existing background object." } ] } } ], "scene_type": "normal", "dependent_scene_index": 12, "dependency_reason": "exact_background from previous shot in the same room (S12_Shot7)", "_shot_index": 20, "visible_entities": [ "P05", "P08", "L07" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 12, "shot_index": 20 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Opposite the bed at eye height, the camera locks into a wide shot (WS), eye-level, perfectly centered on the blank white wall above the bed. The red circular symbol sits in the middle of the frame like wet paint, large but not overwhelming, with the bed forming a low horizontal base and no characters visible.", "lighting_mood": "Flat cold interior light drains the wall to bone-white while the red circle appears saturated and occult, as if projected into reality.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "P05", "P08", "L07" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C05", "outlook_id": "O06" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "bg_daughter_bedroom_night_blood_curtain_body_red_circle", "reference_usage": "exact_background", "owned_objects": [ "backpack", "bed", "circle", "curtain", "doll", "door", "floor", "photo", "pillow", "sheet", "wall", "window" ], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "owned_objects lists all environment objects already drawn in the background reference — do not create new ones", "when focusing on an owned object, anchor it explicitly: 'the door from the reference'" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [ { "scene_index": 12, "shot_index": 7, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C05O06", "policy": "required" }, { "kind": "background", "id": "bg_daughter_bedroom_night_blood_curtain_body_red_circle", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "72e275431cdd6e56", "status": "contract_violation" }, { "scene_index": 13, "heading": "S#13. 옥탑방 외곽 / N", "beat_title": "Reaction - Escaping Horror", "representative_moment": "혜수가 S11 옥탑방에서 도망쳐 나온 수리영(C01)의 어깨를 붙잡는 순간.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Medium shot, eye-level, cold night fog with pulsing red and blue police strobes", "t2i_prompt": "Photorealistic cinematic still. Cold night fog envelops the scene, illuminated by pulsing red and blue police strobe lights. In the foreground, a young East Asian woman in a navy windbreaker, with back to camera, stands slumped. Behind her in the middle ground, C02O03 in a gray leather jacket, a middle-aged East Asian woman, stands at three-quarter angle from the right. C02O03's hands firmly grip the young woman's shoulders across the shared concrete exterior path they both occupy. C02O03's eyes fixed on the young woman. The building door from the reference and the boundary wall recede into the haze, catching the harsh alternating colored glare. Focus on the firm grip bridging the two figures.", "outfit_assignments": [ { "character_id": "C02", "outlook_id": "O03" } ], "source_facts": [ "Scene text mentions sirens sounding and Hye-soo arriving ('동시에 사이렌이 울리고~. 차에서 내린 혜수').", "shot description: 수리영의 어깨를 양손으로 감싸 쥔 혜수의 정지 구도." ], "visual_inferences": [ "fixed_element lighting: police_strobe_lighting with pulsing red and blue glare", "Su-ri-young's back to the camera emphasizing her slumped, defeated posture" ], "creative_decisions": [ "Medium framing with foreground/middle-ground separation to emphasize Hye-soo's supportive grip", "Using the shared concrete path to anchor both figures in the same physical space", "Applying common noun + demographic descriptor for the back-to-camera figure per id_policy" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "446a1980eabb063b", "owned_hash": "da184dd47ed9e9fa", "camera_direction_hash": "7d71cb484b0cd941", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 11, "dependency_reason": "S11 옥탑방 외곽 배경의 연속선상에 있으며, 배경 참조 이미지를 재사용하여 위치적 연속성을 유지", "_shot_index": 4, "visible_entities": [ "C01", "C02", "L05", "L06" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 13, "shot_index": 4 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "On the exterior path between the stairway and police car, the camera holds a medium wide shot (MWS), eye-level, in a locked two-shot after the preceding movement. 혜수's two hands gripping 수리영's shoulders become the center anchor, with 수리영 slightly nearer to camera and 혜수 squared behind her in the middle ground, the siren-lit rooftop receding around them.", "lighting_mood": "Blue-red police strobes rake across cold rooftop fog, freezing the women in alternating panic and procedural glare.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "C02", "L05", "L06" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C02", "outlook_id": "O03" }, { "character_id": "C04", "outlook_id": "O05" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "bg_approach_night_siren_arrival", "reference_usage": "exact_background", "owned_objects": [ "CCTV camera", "boundary wall", "building", "door", "drain channel", "emergency vehicle", "gate", "landing", "plume", "railing", "road", "security light", "stain", "stair", "utility pole", "window", "yard" ], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "owned_objects lists all environment objects already drawn in the background reference — do not create new ones", "when focusing on an owned object, anchor it explicitly: 'the door from the reference'" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "police_strobe_lighting", "element_type": "environment_state", "character_name": "", "description": "The dark exterior environment is enveloped in cold night fog and continuously swept by pulsing red and blue police strobe lights. These harsh, alternating colored glares cut through the mist, illuminating the surroundings with a striking and urgent atmosphere.", "applies_to_shots": [ 4, 5, 6 ], "source_facts": [ "Scene text mentions sirens sounding and a police car arriving ('동시에 사이렌이 울리고~. 차에서 내린 혜수').", "Shot 4 describes 'Blue-red police strobes rake across cold rooftop fog'.", "Shot 5 describes 'pulsing red-blue siren light'.", "Shot 6 mentions 'A cold blue base with intermittent crimson siren flicker'." ], "visual_inferences": [ "The physical police vehicle is off-screen or out of focus, but its strong flashing lights universally dictate the visual tone of the surrounding area." ], "creative_decisions": [ "Consistently applying the alternating blue and red strobe effects across all shots to visually anchor the urgency and shock of the moment." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 11, "shot_index": 6, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C02O03", "policy": "required" }, { "kind": "character_outlook", "id": "C04O05", "policy": "required" }, { "kind": "background", "id": "bg_approach_night_siren_arrival", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "ae9366b7728dc0bd" }, { "scene_index": 13, "heading": "S#13. 옥탑방 외곽 / N", "beat_title": "충격적인 소식을 전하는 수리영", "representative_moment": "두려움에 떠는 수리영(C01)이 혜수(C02)를 올려다보며 충격적인 사실을 털어놓는다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "extreme low angle head-and-shoulder close-up, dramatic police strobe lighting", "t2i_prompt": "Photorealistic cinematic still. Low observer looking up at a steep angle, a head-and-shoulder composition. C01O02 in a navy windbreaker, a young East Asian woman, occupies the lower center, her upper chest visible at the bottom edge, chin lifted trembling in three-quarter left profile, tearful eyes fixed on C02O03. C02O03 in a gray leather jacket, a middle-aged East Asian woman, stands in the extreme foreground with her back to the camera, her neck and jawline skimming the upper frame edge, looking down at C01O02. They share the [L05: dark wet concrete stair landing]. Pulsing blue and red police strobe lighting washes over C01O02's wet cheeks and reflects under C02O03's jaw. Cold night haze and soft red-blue bokeh blur in the deep background behind them.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C02", "outlook_id": "O03" } ], "source_facts": [ "Scene action: '수리영 (올라다보며) 엄마가, 엄마가... 죽어있어.'", "Shot 5 description: 'Low and close in front of 수리영, low angle, looking up so 혜수's chin and lower face skim the upper edge like a looming frame.'" ], "visual_inferences": [ "fixed_element police_strobe_lighting dictates the red and blue glow reflecting on their skin.", "C02 is placed in the extreme foreground with her back to the camera to physically create the 'looming frame' effect described by the DP." ], "creative_decisions": [ "Chose a head-and-shoulder composition to satisfy the face identifiability rule while avoiding a 'floating head' visual error.", "Used soft red-blue bokeh in the depth to maintain focus on C01's emotional expression without distracting background details." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "97a4435142e937ac", "owned_hash": "da184dd47ed9e9fa", "camera_direction_hash": "603a29736687bcab", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 13, "dependency_reason": "previous_shot_refs indicates S13 Shot 4 for the location, though close framing intentionally omits specific background anchors beyond the shared stair landing.", "_shot_index": 5, "visible_entities": [ "C01", "C02", "L05", "L06" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 13, "shot_index": 5 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Low and close in front of 수리영, the camera finishes its inward move as a close-up (CU), low angle, looking up so 혜수's chin and lower face skim the upper edge like a looming frame. 수리영's tearful face occupies the lower center, her eyes lifted toward 혜수 while the police lights smear behind them in soft blue-red bokeh.", "lighting_mood": "Cold night haze is punctured by pulsing red-blue siren light that catches in 수리영's tears and under 혜수's jaw.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "C02", "L05", "L06" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C02", "outlook_id": "O03" }, { "character_id": "C04", "outlook_id": "O05" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_approach_night_siren_arrival", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "police_strobe_lighting", "element_type": "environment_state", "character_name": "", "description": "The dark exterior environment is enveloped in cold night fog and continuously swept by pulsing red and blue police strobe lights. These harsh, alternating colored glares cut through the mist, illuminating the surroundings with a striking and urgent atmosphere.", "applies_to_shots": [ 4, 5, 6 ], "source_facts": [ "Scene text mentions sirens sounding and a police car arriving ('동시에 사이렌이 울리고~. 차에서 내린 혜수').", "Shot 4 describes 'Blue-red police strobes rake across cold rooftop fog'.", "Shot 5 describes 'pulsing red-blue siren light'.", "Shot 6 mentions 'A cold blue base with intermittent crimson siren flicker'." ], "visual_inferences": [ "The physical police vehicle is off-screen or out of focus, but its strong flashing lights universally dictate the visual tone of the surrounding area." ], "creative_decisions": [ "Consistently applying the alternating blue and red strobe effects across all shots to visually anchor the urgency and shock of the moment." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 13, "shot_index": 4, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C02O03", "policy": "required" }, { "kind": "character_outlook", "id": "C04O05", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "f34f36eae14cf9a9" }, { "scene_index": 13, "heading": "S#13. 옥탑방 외곽 / N", "beat_title": "충격적인 소식을 듣는 혜수", "representative_moment": "C02가 충격에 빠진 표정으로 눈을 크게 뜨고 미간을 찌푸린 클로즈업 순간.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Eye-level close-up, clean frontal framing, shallow focus", "t2i_prompt": "Photorealistic cinematic still. C02O03 in a dark leather jacket, a middle-aged Asian woman, facing the camera. Tight head-and-shoulder composition, framing her widened, shocked eyes and furrowed brow dead center, neck and upper chest visible in the frame. Her jaw is tightened, eyes fixed on a person off-camera. Cold blue night fog ambient washes her features, while an intermittent crimson police strobe light reflects on the side of her cheek and shoulder. Focus on her wide eyes reflecting the flashing light. [L06: Out-of-focus dark night fog, a blurred wet silhouette of a rooftop railing in the shallow background behind her].", "outfit_assignments": [ { "character_id": "C02", "outlook_id": "O03" } ], "source_facts": [ "Scene action: '혜수 (놀라) 뭐?!'", "Shot description: 미간을 찌푸린 채 두 눈이 커진 혜수의 충격받은 표정", "DP instructions: reverse cut, close-up, eye-level, clean and frontal, widened eyes, furrowed brow dead center" ], "visual_inferences": [ "fixed_element police_strobe_lighting: cold blue night fog and pulsing crimson police strobe reflections on her face", "shallow focus background with wet rooftop railing to isolate her emotional reaction" ], "creative_decisions": [ "Head-and-shoulder framing chosen over full face-filling frame to maintain realistic body attachment and avoid floating-head artifacts", "Directional crimson strobe reflection emphasizes the sudden, shocking realization without cutting across features" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "be5b2b43c7b38502", "owned_hash": "da184dd47ed9e9fa", "camera_direction_hash": "a2946dae0d2b8b29", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 13, "dependency_reason": "같은 옥탑방 외부 공간의 연속된 순간이며, 환경적으로 짙은 안개와 경찰차 사이렌(파란색/빨간색 교차 불빛)이 연속성을 가진다.", "_shot_index": 6, "visible_entities": [ "C02", "L05", "L06" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 13, "shot_index": 6 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "The reverse cut lands less than a meter from 혜수 in a close-up (CU), eye-level, clean and frontal as an emotional punctuation. Her widened eyes and furrowed brow sit dead center while the dark rooftop behind her falls into shallow focus, leaving only the shock of 수리영's words in the frame.", "lighting_mood": "A cold blue base with intermittent crimson siren flicker sharpens the fear in her eyes and brow.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C02", "L05", "L06" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C02", "outlook_id": "O03" }, { "character_id": "C04", "outlook_id": "O05" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_approach_night_siren_arrival", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "police_strobe_lighting", "element_type": "environment_state", "character_name": "", "description": "The dark exterior environment is enveloped in cold night fog and continuously swept by pulsing red and blue police strobe lights. These harsh, alternating colored glares cut through the mist, illuminating the surroundings with a striking and urgent atmosphere.", "applies_to_shots": [ 4, 5, 6 ], "source_facts": [ "Scene text mentions sirens sounding and a police car arriving ('동시에 사이렌이 울리고~. 차에서 내린 혜수').", "Shot 4 describes 'Blue-red police strobes rake across cold rooftop fog'.", "Shot 5 describes 'pulsing red-blue siren light'.", "Shot 6 mentions 'A cold blue base with intermittent crimson siren flicker'." ], "visual_inferences": [ "The physical police vehicle is off-screen or out of focus, but its strong flashing lights universally dictate the visual tone of the surrounding area." ], "creative_decisions": [ "Consistently applying the alternating blue and red strobe effects across all shots to visually anchor the urgency and shock of the moment." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 13, "shot_index": 5, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C02O03", "policy": "required" }, { "kind": "character_outlook", "id": "C04O05", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "86c25f1147aaed4f" }, { "scene_index": 14, "heading": "S#14. 옥탑방 안 / N", "beat_title": "사라진 시신과 핏자국, 비정상적으로 깨끗해진 방", "representative_moment": "C02 looks at the unnervingly spotless living room with confusion, framed over C04's shoulder.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Eye-level medium over-the-shoulder shot, sickly green-gray fluorescent lighting with a cold blue accent.", "t2i_prompt": "Photorealistic cinematic still. Eye-level medium over-the-shoulder shot from the interior passage. C04O05 in a beige jacket, a middle-aged East Asian man, leans into the doorway, his back and soft shoulder framing the left foreground, side of his face visible, his eyes fixed on the empty living room. Sharing the same room, C02O03 in a gray leather jacket, a young East Asian woman, stands in the middle ground facing the camera, pausing mid-search, her eyes fixed on the empty living room space. Focus on C02O03 centered against the unnervingly clean interior. The living room floor is completely spotless without any marks. The TV from the reference glows dimly, its screen angled partly toward the camera. A casual backpack with a small plush doll rests quietly on the floor near the wall. Sickly fluorescent green-gray light washes across the pristine walls, mixing with a cold blue flicker from the TV.", "outfit_assignments": [ { "character_id": "C02", "outlook_id": "O03" }, { "character_id": "C04", "outlook_id": "O05" } ], "source_facts": [ "김형사(허탈한 표정) 없는데.. 아무 것도.", "지나칠 정도로 깨끗하게 정돈된 방안. 핏자국도 보이지 않고", "shot description: 김형사의 어깨 너머로 텅 빈 거실을 바라보는 혜수의 정면." ], "visual_inferences": [ "The living room is completely spotless, extending the unnerving cleanliness from the bedroom to the entire apartment.", "P09 (backpack) is visible on the floor near the wall as a persistent prop." ], "creative_decisions": [ "Using a medium over-the-shoulder shot to frame C02's reaction against the meticulously cleaned background.", "Applying a sickly fluorescent green-gray ambient light mixed with cold blue TV glow to emphasize the unnatural, eerie state of the room." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "63aeef030ebc4285", "owned_hash": "cd9e08476100c331", "camera_direction_hash": "cc7a00f0b946560d", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "TV", "violating_phrase": "The TV from the reference glows dimly, its screen angled partly toward the camera.", "reason": "This describes the TV as an in-scene object being newly rendered with specific placement and visual details rather than only referencing the existing TV." } ] } } ], "scene_type": "normal", "dependent_scene_index": 5, "dependency_reason": "exact_background ref from S5 S14; requires identical room geometry and owned objects like the TV and AC.", "_shot_index": 2, "visible_entities": [ "C02", "C04", "P09", "L07" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 14, "shot_index": 2 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From inside the rooftop room entry area, the camera holds a medium shot (MS), eye-level, over-the-shoulder behind 김형사 as his soft shoulder frames the left foreground. 혜수 faces the empty living room from the middle ground, centered against an unnervingly clean interior, while the negative space behind her makes the vanished evidence feel louder than any body.", "lighting_mood": "Sickly fluorescent green-gray interior light mixes with the TV's cold blue flicker, emphasizing the impossible cleanliness of the room.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C02", "C04", "P09", "L07" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C02", "outlook_id": "O03" }, { "character_id": "C04", "outlook_id": "O05" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "bg_living_kitchen_night_cleaned_no_cups", "reference_usage": "exact_background", "owned_objects": [ "AC", "TV", "bathroom door", "bedroom door", "chair", "cooktop", "curtain", "dining table", "door", "floor cushion", "framed photo", "kitchen counter", "media console", "refrigerator", "shelf", "shoe cabinet", "shoe step", "sink", "sliding window", "sofa", "storage door" ], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "owned_objects lists all environment objects already drawn in the background reference — do not create new ones", "when focusing on an owned object, anchor it explicitly: 'the door from the reference'" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "spotless_apartment_interior", "element_type": "environment_state", "character_name": "", "description": "The interior of the rooftop apartment, encompassing the living area and bedroom, is impossibly clean and meticulously organized. There are absolutely no traces of blood, no bodies, and no strange marks on any surfaces. The floors, walls, and furniture are eerily spotless under the cold, sickly interior lighting, presenting an unnerving contrast to the expected crime scene.", "applies_to_shots": [ 2, 6, 8 ], "source_facts": [ "지나칠 정도로 깨끗하게 정돈된 방안. 핏자국도 보이지 않고", "식탁 역시 깨끗한", "unnervingly clean interior", "spotless room" ], "visual_inferences": [ "The cleanliness is uniformly maintained across all visible areas (living room, bedroom, dining area) to emphasize the sudden and impossible disappearance of the gruesome scene." ], "creative_decisions": [ "The cold, sickly lighting is used to enhance the eerie feeling of the completely sanitized and empty space." ], "confidence": "high" }, { "element_id": "backpack_with_doll", "element_type": "persistent_prop", "character_name": "", "description": "A casual backpack with a small plush doll dangling from it rests quietly on the floor near the wall in the apartment. It remains completely stationary throughout the search, serving as a silent, slightly displaced artifact of the room's normal state before the anomalous events.", "applies_to_shots": [ 2, 6, 8 ], "source_facts": [ "수리영 방안에 놓여진 백팩, 인형. 약간씩 위치가 달라진.", "등장 요소: P09(인형 달린 백팩) in shots 2, 6, 8" ], "visual_inferences": [ "Placed on the floor near a wall so it can be visible in the background across different framing angles without obstructing the moving characters." ], "creative_decisions": [ "The backpack is used as an eerie anchor to mundane reality in the otherwise impossibly cleaned room." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 5, "shot_index": 14, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C02O03", "policy": "required" }, { "kind": "character_outlook", "id": "C04O05", "policy": "required" }, { "kind": "background", "id": "bg_living_kitchen_night_cleaned_no_cups", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "d342d8f5b2507454", "status": "contract_violation" }, { "scene_index": 14, "heading": "S#14. 옥탑방 안 / N", "beat_title": "사라진 시신과 흔적, 혼란에 빠진 수리영", "representative_moment": "피와 시신이 사라진 깨끗한 방을 보며 파리해진 얼굴로 굳어있는 C01의 클로즈업.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Eye-level head-and-shoulder close-up with shallow depth of field, cold dawn-gray lighting", "t2i_prompt": "Photorealistic cinematic still. Head-and-shoulder composition, holding C01O02 in a navy windbreaker, a young East Asian woman, in the center-right of the frame. She stands at three-quarter angle from the left, head jerked back slightly, her wide eyes fixed on an empty curtain corner. Focus on her drained, pale expression. Cold dawn-gray interior light washes over her face, draining her skin tone, with a faint sickly spill catching her shoulder and jawline, devoid of any warm relief. Behind her, [L07: the room dissolves into a soft blur through shallow depth of field, revealing the edge of a clean, closed-looking curtain and a neatly made bed. On the blurred floor, P09, a casual backpack with a small plush doll, rests silently].", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "얼굴이 파리해지는 수리영. ‘어----?’ ‘분명히 봤는데...’", "지나칠 정도로 깨끗하게 정돈된 방안. 핏자국도 보이지 않고", "수리영 방안에 놓여진 백팩, 인형. 약간씩 위치가 달라진.", "shot description: 수리영의 클로즈업. 피와 시신, 기괴한 표식까지 모두 감쪽같이 사라진 깨끗한 방을 보며 극심한 혼란을 느낌" ], "visual_inferences": [ "fixed_element: spotless_apartment_interior (The interior of the rooftop apartment... eerily spotless under the cold, sickly interior lighting)", "fixed_element: backpack_with_doll (A casual backpack with a small plush doll dangling from it rests quietly on the floor near the wall)", "Head jerked back slightly to emphasize the sudden shock and disbelief." ], "creative_decisions": [ "Shallow depth of field isolates her drained expression while keeping the impossibly clean background elements (bed, curtain, backpack) as soft, eerie evidence of the anomaly.", "Cold dawn-gray light washes the skin, enhancing the sickly, drained horror mood without warm relief.", "Positioned her in the center-right to give breathing room for her gaze toward the empty corner where the body was supposed to be." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "11381d06738cae83", "owned_hash": "a7f71fcf41c4d0e3", "camera_direction_hash": "f35e94b6b36fe3e1", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 5, "dependency_reason": "수리영의 방 내부 환경의 연속성을 유지하되, 현재는 시신과 혈흔이 사라진 상태를 묘사해야 함", "_shot_index": 6, "visible_entities": [ "C01", "P09", "L07" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 14, "shot_index": 6 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Near the foot of the bed at eye level, the camera completes the stage dolly-in into a tight CU with shallow DOF, holding 수리영’s drained face in the center-right of frame while the spotless room dissolves into blur behind her. This is a direct observer view, not her hallucination: the empty curtain corner and clean bed remain only as soft background evidence around her widened eyes.", "lighting_mood": "Cold dawn-gray interior light drains her skin tone, with a faint sickly room spill and no warm relief.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "P09", "L07" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C02", "outlook_id": "O03" }, { "character_id": "C04", "outlook_id": "O05" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_daughter_bedroom_night_cleaned_rearranged", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "spotless_apartment_interior", "element_type": "environment_state", "character_name": "", "description": "The interior of the rooftop apartment, encompassing the living area and bedroom, is impossibly clean and meticulously organized. There are absolutely no traces of blood, no bodies, and no strange marks on any surfaces. The floors, walls, and furniture are eerily spotless under the cold, sickly interior lighting, presenting an unnerving contrast to the expected crime scene.", "applies_to_shots": [ 2, 6, 8 ], "source_facts": [ "지나칠 정도로 깨끗하게 정돈된 방안. 핏자국도 보이지 않고", "식탁 역시 깨끗한", "unnervingly clean interior", "spotless room" ], "visual_inferences": [ "The cleanliness is uniformly maintained across all visible areas (living room, bedroom, dining area) to emphasize the sudden and impossible disappearance of the gruesome scene." ], "creative_decisions": [ "The cold, sickly lighting is used to enhance the eerie feeling of the completely sanitized and empty space." ], "confidence": "high" }, { "element_id": "backpack_with_doll", "element_type": "persistent_prop", "character_name": "", "description": "A casual backpack with a small plush doll dangling from it rests quietly on the floor near the wall in the apartment. It remains completely stationary throughout the search, serving as a silent, slightly displaced artifact of the room's normal state before the anomalous events.", "applies_to_shots": [ 2, 6, 8 ], "source_facts": [ "수리영 방안에 놓여진 백팩, 인형. 약간씩 위치가 달라진.", "등장 요소: P09(인형 달린 백팩) in shots 2, 6, 8" ], "visual_inferences": [ "Placed on the floor near a wall so it can be visible in the background across different framing angles without obstructing the moving characters." ], "creative_decisions": [ "The backpack is used as an eerie anchor to mundane reality in the otherwise impossibly cleaned room." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 5, "shot_index": 9, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C02O03", "policy": "required" }, { "kind": "character_outlook", "id": "C04O05", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "cafde308c8ccb61e" }, { "scene_index": 14, "heading": "S#14. 옥탑방 안 / N", "beat_title": "사라진 시신과 짜증내는 김형사", "representative_moment": "C04가 허리에 손을 얹고 짜증 섞인 표정으로 C01을 내려다보고 있다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Medium wide shot, low angle over-the-shoulder, sickly fluorescent lighting", "t2i_prompt": "Photorealistic cinematic still. Focus on C04O05's accusatory posture across the room. Framed from a low observer position at hip height, the back of the head and shoulder of C01O02 in a navy windbreaker, a young Asian woman, anchor the lower left foreground as an over-the-shoulder element, hunched low and looking up. Occupying the same spotless living area across from her, C04O05 in a beige jacket, a middle-aged Asian man, stands in the center frame at a three-quarter angle from the right, looming with hands on waist, looking down at the foreground figure. His torso and upper body fill the vertical space, head near the top frame edge. Sickly fluorescent ceiling light casts cold shadows across the eerily clean and completely sanitized room. Behind him, the door from the reference stands closed, and the clean dining table from the reference recedes into the background. P09, a casual backpack with a small plush doll, rests quietly on the floor near the wall.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C04", "outlook_id": "O05" } ], "source_facts": [ "김형사 (답답) 허 참! 죽은 사람이 발이 달렸을 리도 없고.", "지나칠 정도로 깨끗하게 정돈된 방안. 핏자국도 보이지 않고", "수리영 방안에 놓여진 백팩, 인형. 약간씩 위치가 달라진." ], "visual_inferences": [ "Suyeong is seen from a low over-the-shoulder angle to make Kim Detective appear looming and accusatory.", "The backpack P09 remains in the background as a subtle anchor to reality." ], "creative_decisions": [ "Medium wide shot with a low framing height to emphasize the power dynamic and Kim's frustration.", "Sickly fluorescent lighting highlights the unnervingly clean environment, contrasting with Suyeong's panic." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "2733781646f713d5", "owned_hash": "cd9e08476100c331", "camera_direction_hash": "765934081edb6589", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "door", "violating_phrase": "the door from the reference stands closed", "reason": "It refers to the door itself in the scene without making it clear as an existing anchor-only mention, which can read as redrawing the owned door." }, { "owned_object": "dining table", "violating_phrase": "the clean dining table from the reference recedes into the background", "reason": "It mentions the dining table as part of the prompt in a way that can imply placing or rendering that owned object rather than only anchoring to what already exists." } ] } } ], "scene_type": "normal", "dependent_scene_index": 14, "dependency_reason": "exact_background ref S14_Shot6: same room, preserving the clean background layout while focusing on a new character interaction moment.", "_shot_index": 8, "visible_entities": [ "C01", "C04", "P09", "L07" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 14, "shot_index": 8 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From low beside 수리영 in the living area, a static MWS low angle frames 김형사 upper-center as he looms over her, his hands-on-waist silhouette cutting into the fluorescent haze. 수리영’s shoulder and bowed head sit low in the left foreground as an OTS pressure point, while the clean dining area recedes behind them.", "lighting_mood": "Sickly police-procedural fluorescent light mixes with cold apartment shadows, making 김형사 feel accusatory and 수리영 smaller.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "C04", "P09", "L07" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C02", "outlook_id": "O03" }, { "character_id": "C04", "outlook_id": "O05" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "bg_living_kitchen_night_cleaned_no_cups", "reference_usage": "exact_background", "owned_objects": [ "AC", "TV", "bathroom door", "bedroom door", "chair", "cooktop", "curtain", "dining table", "door", "floor cushion", "framed photo", "kitchen counter", "media console", "refrigerator", "shelf", "shoe cabinet", "shoe step", "sink", "sliding window", "sofa", "storage door" ], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "owned_objects lists all environment objects already drawn in the background reference — do not create new ones", "when focusing on an owned object, anchor it explicitly: 'the door from the reference'" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "spotless_apartment_interior", "element_type": "environment_state", "character_name": "", "description": "The interior of the rooftop apartment, encompassing the living area and bedroom, is impossibly clean and meticulously organized. There are absolutely no traces of blood, no bodies, and no strange marks on any surfaces. The floors, walls, and furniture are eerily spotless under the cold, sickly interior lighting, presenting an unnerving contrast to the expected crime scene.", "applies_to_shots": [ 2, 6, 8 ], "source_facts": [ "지나칠 정도로 깨끗하게 정돈된 방안. 핏자국도 보이지 않고", "식탁 역시 깨끗한", "unnervingly clean interior", "spotless room" ], "visual_inferences": [ "The cleanliness is uniformly maintained across all visible areas (living room, bedroom, dining area) to emphasize the sudden and impossible disappearance of the gruesome scene." ], "creative_decisions": [ "The cold, sickly lighting is used to enhance the eerie feeling of the completely sanitized and empty space." ], "confidence": "high" }, { "element_id": "backpack_with_doll", "element_type": "persistent_prop", "character_name": "", "description": "A casual backpack with a small plush doll dangling from it rests quietly on the floor near the wall in the apartment. It remains completely stationary throughout the search, serving as a silent, slightly displaced artifact of the room's normal state before the anomalous events.", "applies_to_shots": [ 2, 6, 8 ], "source_facts": [ "수리영 방안에 놓여진 백팩, 인형. 약간씩 위치가 달라진.", "등장 요소: P09(인형 달린 백팩) in shots 2, 6, 8" ], "visual_inferences": [ "Placed on the floor near a wall so it can be visible in the background across different framing angles without obstructing the moving characters." ], "creative_decisions": [ "The backpack is used as an eerie anchor to mundane reality in the otherwise impossibly cleaned room." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 14, "shot_index": 6, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C02O03", "policy": "required" }, { "kind": "character_outlook", "id": "C04O05", "policy": "required" }, { "kind": "background", "id": "bg_living_kitchen_night_cleaned_no_cups", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "45ff7d12b893bef9", "status": "contract_violation" }, { "scene_index": 15, "heading": "S#15. 해안가. 버스정류장 부근 / N", "beat_title": "엄마의 무사를 믿으며 급히 자리를 뜨는 수리영", "representative_moment": "수리영(C01)이 건네받지 않은 캔 음료를 둔 채 벤치에서 몸을 일으키며 안개 낀 도로를 응시한다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "chest-height observer perspective, muted blue-gray coastal night haze", "t2i_prompt": "Photorealistic cinematic still. Captured at chest height from the front-left corner of a damp, weathered bus stop bench. C01O02 in a navy windbreaker, a young East Asian woman, occupies the right third of the frame at a three-quarter angle from the right, caught mid-rise from the bench, her body pushing upward with eyes fixed on the foggy road ahead. At the far left edge of the frame, the extended forearm of an older East Asian woman holds an unopened metallic can drink outward over the shared bench surface, an unaccepted offer. A cold, blue-gray coastal night environment heavily blanketed in sea mist surrounds them. The bench surface and the metallic can faintly catch weak specular light from a sparse roadside streetlamp. The empty coastal road stretches into the mist behind the rising figure. A faded bus stop sign, its face angled partly toward the camera, stands in the dim background. Focus on the tense upward momentum of C01O02 leaving the bench, contrasting with the still, offered can in the muted blue-gray haze.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "S#15 action: '수리영 (훅 일어나며) 집에 가봐야 겠어.'", "shot description: 'At chest height from the front-left corner of the bus-stop bench, a held MS eye-level composition lets 수리영’s sudden rise become the only violent motion in frame.'", "shot description: 'The offered can and 혜수’s partial arm remain at the left edge'" ], "visual_inferences": [ "fixed_element coastal_night_atmosphere: 'A cold, blue-gray coastal night environment heavily blanketed in sea mist.'", "fixed_element damp_bus_stop_bench: 'A standard roadside bus stop bench... Its damp surface faintly catches and reflects the weak specular light'", "혜수 is an unregistered extra not allowed as a C## entity, represented here as an older East Asian woman's forearm extending the can." ], "creative_decisions": [ "Chest-height observer framing anchors the shared bench surface while separating the dynamic upward motion of C01 from the static extended arm on the left edge.", "Muted blue-gray color temperature and weak specular reflections emphasize the damp, cold, and anxious mood of the coastal night." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "07581b9b47838707", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "a89375ffbf9ffefd", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "Not directly dependent on a prior full scene's specific action, relies on its own fixed_elements continuity.", "_shot_index": 3, "visible_entities": [ "C01", "L14" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 15, "shot_index": 3 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At chest height from the front-left corner of the bus-stop bench, a held MS eye-level composition lets 수리영’s sudden rise become the only violent motion in frame. The offered can and 혜수’s partial arm remain at the left edge, while 수리영 lifts from the bench on the right third and turns toward the foggy road.", "lighting_mood": "Muted blue-gray coastal night haze flattens the scene, with weak streetlight glinting off the damp bench and can.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "L14" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C02", "outlook_id": "O03" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "not_applicable", "bg_id": null, "reference_usage": "none", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "no chain_bg is bound to this shot — describe the location freely without referencing a background image", "do not use 'from the reference' / 'the existing room' / 'preserving the room' style phrasing — there is no reference image to refer to" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "coastal_night_atmosphere", "element_type": "environment_state", "character_name": "", "description": "A cold, blue-gray coastal night environment heavily blanketed in sea mist. The area is dimly lit by sparse, weak roadside streetlights, creating a desaturated, moody, and damp overall atmosphere.", "applies_to_shots": [ 3, 6 ], "source_facts": [ "Muted blue-gray coastal night haze flattens the scene", "weak streetlight glinting", "Cold sea mist and sparse roadside light create a predatory, desaturated image" ], "visual_inferences": [ "The heavy mist diffuses the artificial street lighting, significantly reducing background visibility and color saturation." ], "creative_decisions": [ "Maintain a consistent cool color palette (blue-gray) and foggy diffusion across both framings to anchor the spatial continuity of the coastal road." ], "confidence": "high" }, { "element_id": "damp_bus_stop_bench", "element_type": "persistent_prop", "character_name": "", "description": "A standard roadside bus stop bench situated near the misty road. Its damp surface faintly catches and reflects the weak specular light from the nearby streetlamp.", "applies_to_shots": [ 3, 6 ], "source_facts": [ "weak streetlight glinting off the damp bench", "front-left corner of the bus-stop bench", "stays near the bus stop in the midground" ], "visual_inferences": [ "The bench remains a stationary physical anchor point in the environment, marking the exact location where the interaction began." ], "creative_decisions": [ "Ensure the bench has a noticeable sheen from the coastal moisture to enhance the tactile, cold feeling of the night setting." ], "confidence": "high" } ], "previous_shot_refs": [], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C02O03", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "665c40ed46b479e5" }, { "scene_index": 15, "heading": "S#15. 해안가. 버스정류장 부근 / N", "beat_title": "누군가의 감시 속에 헤어지는 두 사람", "representative_moment": "수풀 속 숨겨진 시점이, 벤치 옆에 선 혜수(C02)와 급히 멀어지는 수리영(C01)의 뒷모습을 몰래 지켜본다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "handheld low-angle wide shot from a voyeuristic POV, desaturated blue-gray tones", "t2i_prompt": "Photorealistic cinematic still. From a low hidden position inside the coastal bushes, a handheld low-angle wide shot. Dark wet leaves partially occlude the lower and side edges of the frame. A cold, blue-gray coastal night environment heavily blanketed in sea mist sets a desaturated, moody atmosphere, dimly lit by sparse roadside streetlights. In the midground, the damp surface of a bus stop bench faintly catches the specular light. Beside the bench, C02O03 in a gray leather jacket, a middle-aged Asian woman, stands with her back mostly to the camera but head turned in partial profile, one arm raised toward the road. Farther down the shared foggy coastal road, C01O02 in a navy windbreaker, a young Asian woman, is caught mid-stride facing away, her head turned slightly showing her jawline, one leg lifted in mid-run, jacket flap streaming opposite to motion. Focus on the physical space between the two figures along the hazy path, capturing the predatory tension of the surveillance view.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C02", "outlook_id": "O03" } ], "source_facts": [ "shot description: From a low hidden position inside the coastal bushes, a handheld WS low angle uses through-object framing as leaves partially occlude the lower and side edges", "shot description: This is an unidentified voyeur’s direct first-person surveillance view", "shot description: 혜수 stays near the bus stop in the midground while 수리영 is caught farther away mid-stride, back to camera, one leg lifted toward the road." ], "visual_inferences": [ "fixed_element coastal_night_atmosphere: A cold, blue-gray coastal night environment heavily blanketed in sea mist.", "fixed_element damp_bus_stop_bench: A standard roadside bus stop bench situated near the misty road, damp surface catching light.", "Characters' heads are turned slightly in partial profile to satisfy the face identifiability rule while maintaining the directed back-to-camera stance.", "Jacket flap streams opposite to motion to freeze Suriyoung's running momentum." ], "creative_decisions": [ "Use dark wet leaves around the frame edges to establish the hidden, predatory voyeuristic perspective.", "Bind the figures together spatially by explicitly placing them along the 'shared foggy coastal road' to prevent disjointed rendering." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "2f71246b94f67dad", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "e5d19daff9af5dfb", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 15, "dependency_reason": "동일한 해안가 버스정류장 배경 및 안개 낀 야간의 조명·대기 상태(coastal_night_atmosphere) 연속성 유지", "_shot_index": 6, "visible_entities": [ "C01", "C02", "L14" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 15, "shot_index": 6 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From a low hidden position inside the coastal bushes, a handheld WS low angle uses through-object framing as leaves partially occlude the lower and side edges of the image. This is an unidentified voyeur’s direct first-person surveillance view: 혜수 stays near the bus stop in the midground while 수리영 is caught farther away mid-stride, back to camera, one leg lifted toward the road.", "lighting_mood": "Cold sea mist and sparse roadside light create a predatory, desaturated image with dark leaf silhouettes in the foreground.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "C02", "L14" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C02", "outlook_id": "O03" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "not_applicable", "bg_id": null, "reference_usage": "none", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "no chain_bg is bound to this shot — describe the location freely without referencing a background image", "do not use 'from the reference' / 'the existing room' / 'preserving the room' style phrasing — there is no reference image to refer to" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "coastal_night_atmosphere", "element_type": "environment_state", "character_name": "", "description": "A cold, blue-gray coastal night environment heavily blanketed in sea mist. The area is dimly lit by sparse, weak roadside streetlights, creating a desaturated, moody, and damp overall atmosphere.", "applies_to_shots": [ 3, 6 ], "source_facts": [ "Muted blue-gray coastal night haze flattens the scene", "weak streetlight glinting", "Cold sea mist and sparse roadside light create a predatory, desaturated image" ], "visual_inferences": [ "The heavy mist diffuses the artificial street lighting, significantly reducing background visibility and color saturation." ], "creative_decisions": [ "Maintain a consistent cool color palette (blue-gray) and foggy diffusion across both framings to anchor the spatial continuity of the coastal road." ], "confidence": "high" }, { "element_id": "damp_bus_stop_bench", "element_type": "persistent_prop", "character_name": "", "description": "A standard roadside bus stop bench situated near the misty road. Its damp surface faintly catches and reflects the weak specular light from the nearby streetlamp.", "applies_to_shots": [ 3, 6 ], "source_facts": [ "weak streetlight glinting off the damp bench", "front-left corner of the bus-stop bench", "stays near the bus stop in the midground" ], "visual_inferences": [ "The bench remains a stationary physical anchor point in the environment, marking the exact location where the interaction began." ], "creative_decisions": [ "Ensure the bench has a noticeable sheen from the coastal moisture to enhance the tactile, cold feeling of the night setting." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 15, "shot_index": 3, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C02O03", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "cdaa3cfd8efebe9a" }, { "scene_index": 16, "heading": "S#16. 경찰서 사무실 / N", "beat_title": "위치 추적 요구", "representative_moment": "혜수(C02)가 불길한 예감에 김형사(C04)의 모니터 앞으로 민숙의 전화번호가 찍힌 핸드폰을 내미는 순간.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "tight CU slight Dutch angle at desk height, sickly green-blue fluorescent grading with shallow depth of field", "t2i_prompt": "Photorealistic cinematic still. Shot at desk height with a slight Dutch angle, tight close-up framing. Focus on a smartphone held forward over an office desk, the phone screen facing the camera and occupying two-fifths of the frame, displaying a contact number. Entering from the left edge, C02O03 in a grey leather jacket, a young Asian woman in three-quarter right, her shoulder and profile partially visible, hand gripping the device, eyes fixed on the phone screen. Across the shared desk in the soft out-of-focus background, C04O05 in a beige field jacket, a middle-aged Asian man in three-quarter left, hunched behind a glowing computer monitor, looking down at the phone. Sickly fluorescent office light reflects green-blue off the phone glass, while the blurred monitor adds a cold bureaucratic glow. [L15: Immediate desk surface covered with out-of-focus papers under harsh cool ambient light.]", "outfit_assignments": [ { "character_id": "C02", "outlook_id": "O03" }, { "character_id": "C04", "outlook_id": "O05" } ], "source_facts": [ "Beat action: Hye-su shows Min-suk's saved number on her phone to Kim, asking for immediate location tracking.", "DP camera_direction: tight CU slight Dutch angle frames Hye-su's phone screen occupying two-fifths of the frame.", "Lighting: Sickly fluorescent office light reflects green-blue off the phone glass, with monitor adding bureaucratic glow." ], "visual_inferences": [ "fixed_element police_office_night_environment: the desk acts as a central workspace and physical barrier.", "C02's shoulder and profile are included at the frame edge alongside her gripping fingers to avoid violating the isolated body-part rule when using composite IDs." ], "creative_decisions": [ "Used shallow depth of field to keep the smartphone sharp in the foreground while rendering C04 as a soft, skeptical presence behind the glowing monitor in the background.", "Established 'the shared desk' as the foreground-background spatial anchor." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "247a71dfc42c1509", "owned_hash": "6aa2365bedec5d16", "camera_direction_hash": "e1ff54441110edbe", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "No direct visual dependency on a prior scene specified.", "_shot_index": 4, "visible_entities": [ "C02", "C04", "P05", "L15" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 16, "shot_index": 4 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At desk height directly in front of 김형사’s monitor, a tight CU slight Dutch angle frames 혜수’s phone screen as a through-device insert, the display occupying just under two-fifths of the frame with the monitor plane blurred behind it. 혜수’s sleeve and gripping fingers enter from the left, while 김형사 is only a soft, skeptical presence beyond the screen.", "lighting_mood": "Sickly fluorescent office light reflects green-blue off the phone glass, with the monitor adding a cold bureaucratic glow.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C02", "C04", "P05", "L15" ], "allowed_outlook_pairs": [ { "character_id": "C02", "outlook_id": "O03" }, { "character_id": "C04", "outlook_id": "O05" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_office_night_duty_normal", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "police_office_night_environment", "element_type": "environment_state", "character_name": "", "description": "The interior of a police station office at night, heavily illuminated by harsh, sickly green-blue fluorescent lights. A standard office desk equipped with a softly glowing computer monitor serves as the central workspace, creating a cold, bureaucratic, and airless atmosphere.", "applies_to_shots": [ 4, 6 ], "source_facts": [ "Sickly fluorescent office light reflects green-blue", "monitor adding a cold bureaucratic glow", "Harsh fluorescent green-blue light makes the police office feel airless", "desk line traps the space between them" ], "visual_inferences": [ "The desk is a typical police station night-duty setup with a monitor acting as a physical barrier between the detective and the civilian" ], "creative_decisions": [ "Consistent green-blue color grading across both shots to emphasize the tense and unhealthy night-shift environment" ], "confidence": "high" } ], "previous_shot_refs": [], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C02O03", "policy": "required" }, { "kind": "character_outlook", "id": "C04O05", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "fbd21d09aa38eaf4" }, { "scene_index": 16, "heading": "S#16. 경찰서 사무실 / N", "beat_title": "불길함을 감지하고 위치 추적을 요구하는 혜수", "representative_moment": "혜수가 책상을 짚은 채 심각한 표정으로 김형사에게 휴대폰을 보여주며 다급하게 위치 추적을 요구한다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Low standing height MWS low angle, OTS framing", "t2i_prompt": "Photorealistic cinematic still. Shot from a low standing height, a medium wide low-angle composition. C02O03 in a dark leather jacket, a young Asian woman, leans heavily forward into the center frame at a three-quarter left angle, her upper body and head visible. Both her hands are planted hard on the edge of the desk from the reference, one hand gripping P05, her eyes fixed on the man. In the right foreground, C04O05 in a beige jumper, a middle-aged Asian man, is captured in an over-the-shoulder framing, slumping in the chair from the reference with his back to the camera, his profile partially visible, his eyes fixed on her. The desk from the reference and the softly glowing monitor from the reference act as a shared surface anchor trapping the space between them. Harsh green-blue fluorescent light from the reference washes across her jaw and hands, reflecting on the desk and creating a cold, airless bureaucratic atmosphere. Focus on her tense posture leaning over the desk.", "outfit_assignments": [ { "character_id": "C02", "outlook_id": "O03" }, { "character_id": "C04", "outlook_id": "O05" } ], "source_facts": [ "S#16 action: 혜수 leaning over the desk and showing 민숙's number, asking 김형사 to trace it.", "shot description: 혜수 braces her hands on the desk edge; 김형사 sits in the chair opposite as an OTS counterweight." ], "visual_inferences": [ "fixed_element police_office_night_environment: The desk and monitor from the exact background reference form a physical barrier dividing the two figures.", "혜수's forward posture conveys urgent pressure, emphasized by her hands planted on the desk." ], "creative_decisions": [ "Low standing height MWS angle to emphasize 혜수's dominance in the frame and the physical pressure against the desk.", "OTS framing on 김형사 anchors the spatial relationship while keeping focus on 혜수.", "Harsh green-blue fluorescent light washes across 혜수's skin and the desk surface to enhance the tense, airless mood without slicing features." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "c5c49c027f99d749", "owned_hash": "6aa2365bedec5d16", "camera_direction_hash": "c150bd7ec7c581df", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "desk", "violating_phrase": "the edge of the desk from the reference", "reason": "Although it references the existing desk, the phrase places the subject's hands on it in a way that does not request a new desk; however, the prompt does not indicate adding a new desk." }, { "owned_object": "monitor", "violating_phrase": "the softly glowing monitor from the reference", "reason": "This references the existing monitor rather than asking to create a new one, so it is not a redraw instruction." } ] } } ], "scene_type": "normal", "dependent_scene_index": 16, "dependency_reason": "exact_background from prior shot S16_Shot4", "_shot_index": 6, "visible_entities": [ "C02", "C04", "P05", "L15" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 16, "shot_index": 6 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Across the desk from 혜수 at low standing height, a static MWS low angle holds her leaning into the frame, both hands planted hard on the desk edge in the foreground. 김형사 sits partly right of frame as an OTS counterweight, while the desk line traps the space between them like an interrogation barrier.", "lighting_mood": "Harsh fluorescent green-blue light makes the police office feel airless, sharpening the tension in 혜수’s hands and jaw.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C02", "C04", "P05", "L15" ], "allowed_outlook_pairs": [ { "character_id": "C02", "outlook_id": "O03" }, { "character_id": "C04", "outlook_id": "O05" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "bg_office_night_duty_normal", "reference_usage": "exact_background", "owned_objects": [ "PC", "blind", "box", "chair", "coat rack", "copier", "corridor", "desk", "door", "evidence board", "file cabinet", "fluorescent light", "map board", "monitor", "partition", "phone", "printer", "table", "window" ], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "owned_objects lists all environment objects already drawn in the background reference — do not create new ones", "when focusing on an owned object, anchor it explicitly: 'the door from the reference'" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "police_office_night_environment", "element_type": "environment_state", "character_name": "", "description": "The interior of a police station office at night, heavily illuminated by harsh, sickly green-blue fluorescent lights. A standard office desk equipped with a softly glowing computer monitor serves as the central workspace, creating a cold, bureaucratic, and airless atmosphere.", "applies_to_shots": [ 4, 6 ], "source_facts": [ "Sickly fluorescent office light reflects green-blue", "monitor adding a cold bureaucratic glow", "Harsh fluorescent green-blue light makes the police office feel airless", "desk line traps the space between them" ], "visual_inferences": [ "The desk is a typical police station night-duty setup with a monitor acting as a physical barrier between the detective and the civilian" ], "creative_decisions": [ "Consistent green-blue color grading across both shots to emphasize the tense and unhealthy night-shift environment" ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 16, "shot_index": 4, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C02O03", "policy": "required" }, { "kind": "character_outlook", "id": "C04O05", "policy": "required" }, { "kind": "background", "id": "bg_office_night_duty_normal", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "600beb02c175690c", "status": "contract_violation" }, { "scene_index": 17, "heading": "S#17. 옥탑방 / 동틀무렵", "beat_title": "새벽 안개 속 옥탑방의 정적", "representative_moment": "푸른 새벽 안개 너머로 희미한 노란 불빛이 새어나오는 옥탑방 창문.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Extreme wide shot from a neighboring rooftop at eye level, cold blue dawn haze contrasting with warm yellow light", "t2i_prompt": "Photorealistic cinematic still. Extreme wide shot from a neighboring rooftop at eye level. A pre-sunrise dawn setting enveloped in a cold, dim blue fog. In the foreground, the dark, damp concrete parapet from the reference runs across the frame as empty lines. Across a shadowed, empty narrow alley, a small rooftop structure stands. Focus on the single rectangular window from the reference, emitting a faint, warm yellow glow. The warm light is suspended in a sparse dawn-gray composition, sitting slightly above center, contrasting sharply with the thick pale blue-gray atmospheric veil, creating a quiet occult unease. No people are present.", "outfit_assignments": [], "source_facts": [ "Scene text: 새벽녘 (dawn)", "Shot description: observer at neighboring rooftop, looking at the lit window across the alley" ], "visual_inferences": [ "fixed_element dawn_rooftop_atmosphere: cold dim blue fog, textured concrete parapet, faint warm yellow glow", "The character C12 is removed as requested, leaving the environment devoid of people", "The cat is handled as an owned object in the background reference, so new figures are not added" ], "creative_decisions": [ "Extreme wide shot framing with empty parapet lines in the foreground to emphasize stillness and the isolation of the lit window", "Strong color contrast between the cool ambient fog and the single warm light source" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "d3a0bb7d514f9662", "owned_hash": "c2336a1fc231169c", "camera_direction_hash": "ec4197214fc553fa", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "window", "violating_phrase": "Focus on the single rectangular window from the reference, emitting a faint, warm yellow glow.", "reason": "The prompt specifically asks to depict the window as a visible glowing subject, which is a detailed re-rendering of an owned background object rather than just referencing it." } ] } } ], "scene_type": "normal", "dependent_scene_index": 4, "dependency_reason": "exact_background usage carrying over the established rooftop environment", "_shot_index": 1, "visible_entities": [ "L04" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 17, "shot_index": 1 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From a neighboring rooftop at eye level, a static EWS establishes the rooftop room window as a small warm rectangle suspended in a sparse dawn-gray composition across the alley. The lit window sits slightly above center, with empty parapet lines and haze taking most of the frame in a long take implied stillness.", "lighting_mood": "Dim pre-sunrise blue haze surrounds the faint yellow window glow, creating a quiet occult unease.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "L04" ], "allowed_outlook_pairs": [ { "character_id": "C12", "outlook_id": "O00" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "bg_rooftop_dawn_window_lit_cat", "reference_usage": "exact_background", "owned_objects": [ "cat", "door", "laundry line", "parapet", "post", "roof drain", "roof terrace", "stairwell headhouse", "wall light", "water tank", "window" ], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "owned_objects lists all environment objects already drawn in the background reference — do not create new ones", "when focusing on an owned object, anchor it explicitly: 'the door from the reference'" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "dawn_rooftop_atmosphere", "element_type": "environment_state", "character_name": "", "description": "A pre-sunrise dawn setting enveloped in a cold, dim blue fog. A textured concrete parapet wall runs across the space. In the distance, a small rooftop structure features a single rectangular window that emits a faint, warm yellow glow, contrasting sharply with the cold surrounding haze.", "applies_to_shots": [ 1, 3 ], "source_facts": [ "Scene text: 새벽녘 (dawn)", "Shot 1: Dim pre-sunrise blue haze surrounds the faint yellow window glow, empty parapet lines", "Shot 3: cold dawn rim light, dawn fog as soft peripheral detail, wall texture, window reflection burns as a tiny sick yellow point" ], "visual_inferences": [ "The parapet wall is made of rough, aged concrete typical for older urban villa rooftops in Korea", "The yellow light from the window is the only warm light source in the predominantly cold blue environment" ], "creative_decisions": [ "Maintain the stark color contrast between the ambient cold blue dawn fog and the warm yellow window light as the primary visual anchor across different framing scales" ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 4, "shot_index": 1, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C12O00", "policy": "required" }, { "kind": "background", "id": "bg_rooftop_dawn_window_lit_cat", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "bfa46be608101e05", "status": "contract_violation" }, { "scene_index": 17, "heading": "S#17. 옥탑방 / 동틀무렵", "beat_title": "불 켜진 옥탑방을 응시하는 검은 고양이", "representative_moment": "어두운 새벽 안개 속, 노란 창문 빛이 눈동자에 반사된 검은 고양이의 눈 클로즈업.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Very low worm's eye angle, cold blue dawn rim light with a sharp warm yellow catchlight reflection", "t2i_prompt": "Photorealistic cinematic still. Focus on the single glossy eye of a black cat in right profile, occupying the center frame from a very low upward angle. A tiny, warm yellow reflection of a rectangular lit window burns sharply inside the wet, dark pupil. Cold blue dawn rim light outlines the surrounding black fur. The bottom edge of the frame reveals [L04: rough damp concrete parapet surface], with the peripheral background softening into a dim, cold blue-gray dawn fog.", "outfit_assignments": [], "source_facts": [ "Scene text: 새벽녘, 검은 고양이가 불이 켜져 있는 옥탑 방을 슬쩍 보더니", "Shot description: Very low on the wall, worm's eye reflection shot aligned with one eye of 검은 고양이", "Shot description: tiny catchlight of the lit rooftop window trembles inside it" ], "visual_inferences": [ "fixed_element dawn_rooftop_atmosphere: cold, dim blue fog and a textured concrete parapet wall", "The eye acts as a mirror surface, capturing the only warm light source in the scene", "Worm's eye view places the concrete surface at the lower edge of the frame" ], "creative_decisions": [ "Applied body-part focus rule by using the common noun 'a black cat' instead of C## ID to prevent face-reference distortion", "Replaced forbidden shot-type keywords (ECU, close-up) with spatial framing description ('occupying the center frame from a very low upward angle')", "Contrasted the cold blue rim light against the warm yellow reflection to emphasize the eerie, observing mood" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "da364e9ebe945e03", "owned_hash": "c2336a1fc231169c", "camera_direction_hash": "be5ea07316c2ea43", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 17, "dependency_reason": "동일한 옥탑방 앞의 새벽 환경과 조명 상태(푸른 안개, 노란 창문 빛)를 유지하기 위해 씬 17 앞선 샷의 분위기를 참조함", "_shot_index": 3, "visible_entities": [ "C12", "L04" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 17, "shot_index": 3 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Very low on the wall, the dolly-in resolves into an ECU worm's eye reflection shot aligned with one eye of 검은 고양이. The glossy eye occupies most of the human-scale close-up allowance while a tiny catchlight of the lit rooftop window trembles inside it, leaving fur, wall texture, and dawn fog as soft peripheral detail.", "lighting_mood": "A cold dawn rim light outlines the cat’s black fur while the window reflection burns as a tiny sick yellow point in the eye.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C12", "L04" ], "allowed_outlook_pairs": [ { "character_id": "C12", "outlook_id": "O00" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_rooftop_dawn_window_lit_cat", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "dawn_rooftop_atmosphere", "element_type": "environment_state", "character_name": "", "description": "A pre-sunrise dawn setting enveloped in a cold, dim blue fog. A textured concrete parapet wall runs across the space. In the distance, a small rooftop structure features a single rectangular window that emits a faint, warm yellow glow, contrasting sharply with the cold surrounding haze.", "applies_to_shots": [ 1, 3 ], "source_facts": [ "Scene text: 새벽녘 (dawn)", "Shot 1: Dim pre-sunrise blue haze surrounds the faint yellow window glow, empty parapet lines", "Shot 3: cold dawn rim light, dawn fog as soft peripheral detail, wall texture, window reflection burns as a tiny sick yellow point" ], "visual_inferences": [ "The parapet wall is made of rough, aged concrete typical for older urban villa rooftops in Korea", "The yellow light from the window is the only warm light source in the predominantly cold blue environment" ], "creative_decisions": [ "Maintain the stark color contrast between the ambient cold blue dawn fog and the warm yellow window light as the primary visual anchor across different framing scales" ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 17, "shot_index": 1, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C12O00", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "afd71c934e4b8288" }, { "scene_index": 18, "heading": "S#18. 안방 안 / D", "beat_title": "고양이 소리에 잠에서 깬 수리영", "representative_moment": "C01이 고양이 소리에 번쩍 눈을 뜨고 고개를 치켜든 채 방문 쪽을 바라보는 순간.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "low angle head-and-shoulder composition, cold and pale daylight with a sickly green-gray tint", "t2i_prompt": "Photorealistic cinematic still. Low angle looking up from the bedside, head-and-shoulder composition. C01O02 in a navy windbreaker, a young East Asian woman, positioned at a three-quarter angle from the right. Head tilted up abruptly in mid-motion, hair shifting slightly backward, eyes wide and fixed off-frame toward the unseen doorway. Focus on her startled, pallid expression. Thin cold daylight washes the side of her face with a pale, sickly green-gray ambient tint, highlighting her undercurrent of sleepless exhaustion. [L07: blurred dim headboard close behind her, rumpled bedding occupying the lower foreground frame edge].", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "Shot 2 description: 수리영 snapping awake and lifting her head.", "Shot 2 description: Low at the bedside from the foot side of the bed.", "Shot 2 description: Thin daylight seeps in cold and gray, leaving her face pallid with a faint undercurrent of sleepless exhaustion." ], "visual_inferences": [ "fixed_element cold_gray_interior_lighting: pervasive sickly green-gray tone over the surroundings.", "mid-motion hair shift from sudden waking movement.", "close framing tightens around the headboard and bedding, isolating her startled reaction." ], "creative_decisions": [ "low angle composition to emphasize the sudden jolt of waking up.", "sickly green-gray wash on the face to visually connect the psychological horror theme and exhaustion." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "f3c376d49b531310", "owned_hash": "eb0ef19840d12bce", "camera_direction_hash": "f0799f30d183ac45", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 14, "dependency_reason": "S14_Shot6 is a location reference, but due to skipped_close_framing, the background reference is omitted and only immediate frame-edge surroundings are described.", "_shot_index": 2, "visible_entities": [ "C01", "P05", "P09", "P10", "L07" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 18, "shot_index": 2 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Low at the bedside from the foot side of the bed, the camera finishes an abrupt dolly-in into a CU low angle on 수리영 snapping awake and lifting her head. Her face lands center frame with the headboard and bedding falling out of focus behind her, turning the cat’s cry into a physical jolt rather than a visible source.", "lighting_mood": "Thin daylight seeps in cold and gray, leaving her face pallid with a faint undercurrent of sleepless exhaustion.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "P05", "P09", "P10", "L07" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_master_bedroom_dawn_waiting_empty_home", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "cold_gray_interior_lighting", "element_type": "environment_state", "character_name": "", "description": "A pervasive cold and gray lighting condition that fills the interior spaces. The light casts a pale, sickly, and slightly green-gray tone over the surroundings, creating an eerie, unnatural, and chilling atmosphere across the different rooms.", "applies_to_shots": [ 2, 12, 15 ], "source_facts": [ "Shot 2: 'Thin daylight seeps in cold and gray'", "Shot 12: 'Cold fluorescent bathroom light turns the glass green-gray'", "Shot 15: 'Cold gray daylight mixes with a sickly interior cast'" ], "visual_inferences": [ "A unified desaturated color palette with cold tones to visually connect the different rooms (bedroom, bathroom, dining area) throughout the sequence." ], "creative_decisions": [ "Applying a sickly green-gray tint to the overall ambient light to foreshadow and emphasize the character's psychological instability and the supernatural hallucinations." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 14, "shot_index": 6, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "2792d0d7842d41bd" }, { "scene_index": 18, "heading": "S#18. 옥탑방 내부 (욕실) / D", "beat_title": "거울 속 붉은 원형 표식의 환영", "representative_moment": "거울 표면에 떠오른 붉은 원형 표식의 클로즈업.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Close-up Dutch angle, face-height, cold green-gray fluorescent lighting with sharp crimson accent", "t2i_prompt": "Photorealistic cinematic still. Close-up Dutch angle locked onto a slightly wet, streaked bathroom mirror surface. A sharp, bright crimson circular mark appears clearly on the glass, occupying the central third of the frame. Focus on the crisp edges of the unnatural red sign floating on the reflection. At the lower frame boundary, a partial slice of a wet, white ceramic sink edge is visible. Behind the mark, the reflection of [L07: pale, damp bathroom tiles under cold fluorescent lighting] appears ghosted and slightly out of focus. A pervasive sickly green-gray tint washes the ambient space, heavily contrasting with the intense red of the mark. No human figures or faces are visible in the reflection.", "outfit_assignments": [], "source_facts": [ "S18 action: '거울에 비친 욕실 벽에 갑자기 또 발생하는 붉은 원.'", "shot description: 욕실 거울 표면 위로 선명하게 떠오른 붉은 원형 표식의 클로즈업" ], "visual_inferences": [ "fixed_element: cold_gray_interior_lighting casting a sickly, slightly green-gray tone", "The character's reflection is completely out of frame to emphasize the isolated supernatural sign", "The sink edge provides a lower anchor for the close-up composition" ], "creative_decisions": [ "Dutch angle framing to reflect psychological instability and the hallucinatory nature of the moment", "High color contrast between the unnatural crimson mark and the sickly green-gray fluorescent tiles", "Using shallow depth of field (ghosted background) to isolate the red sign on the mirror surface" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "460266a2f2372176", "owned_hash": "a65185601a83e11d", "camera_direction_hash": "a6c42418a86b5ee0", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 18, "dependency_reason": "동일 씬 앞 샷(Shot 2)의 욕실 환경 및 차가운 조명 톤 연속성을 유지하되, 배경 레퍼런스는 close framing 정책에 따라 합성 과정에서 제외됨.", "_shot_index": 12, "visible_entities": [ "P05", "P09", "P10", "L07" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 18, "shot_index": 12 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At face height, nearly perpendicular to the bathroom mirror, a static CU Dutch angle locks onto the glass surface as a crimson circular mark appears in the central third of frame. This is a hallucination seen via reflection: 수리영’s reflection is reduced out of frame, leaving the sink edge below and tiled wall ghosted behind the impossible red sign.", "lighting_mood": "Cold fluorescent bathroom light turns the glass green-gray while the red circle cuts through as a sick, supernatural accent.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "P05", "P09", "P10", "L07" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_bathroom_dawn_mirror_red_circle_hallucination", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "cold_gray_interior_lighting", "element_type": "environment_state", "character_name": "", "description": "A pervasive cold and gray lighting condition that fills the interior spaces. The light casts a pale, sickly, and slightly green-gray tone over the surroundings, creating an eerie, unnatural, and chilling atmosphere across the different rooms.", "applies_to_shots": [ 2, 12, 15 ], "source_facts": [ "Shot 2: 'Thin daylight seeps in cold and gray'", "Shot 12: 'Cold fluorescent bathroom light turns the glass green-gray'", "Shot 15: 'Cold gray daylight mixes with a sickly interior cast'" ], "visual_inferences": [ "A unified desaturated color palette with cold tones to visually connect the different rooms (bedroom, bathroom, dining area) throughout the sequence." ], "creative_decisions": [ "Applying a sickly green-gray tint to the overall ambient light to foreshadow and emphasize the character's psychological instability and the supernatural hallucinations." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 18, "shot_index": 2, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "a72aaf112faad1bd" }, { "scene_index": 18, "heading": "S#18. 안방 안 / D", "beat_title": "식탁 액자 속 엄마 얼굴이 녹아내리는 끔찍한 환영", "representative_moment": "거실 식탁 위 액자 속 40대 여성의 얼굴이 기괴하게 뭉개지며 녹아내린다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Subjective POV high angle close-up on photograph, cold gray daylight with sickly green-gray interior cast and sharp crimson-brown accents on the melted face.", "t2i_prompt": "Photorealistic cinematic still. High angle close-up, subjective POV. Focus on an upright framed photograph resting on a [L07: clean, cold, and bare dining table surface] occupying the lower frame. Inside the glass frame, a photograph of a middle-aged East Asian woman facing the camera; her face is grotesquely melting and liquefying downwards, turning into a hallucinated surface event with unnatural, sharp crimson-brown streaks. A half-full drinking glass with faint condensation sits heavily blurred in the left foreground. A pervasive cold gray daylight mixes with a sickly green-gray interior cast, washing over the table.", "outfit_assignments": [], "source_facts": [ "Shot 15 description: subjective POV from Suri-young at dining-table height looking at the framed photograph", "mother's portrait appears to liquefy behind the glass, turning a family snapshot into a hallucinated surface event", "drinking glass blurred at the left foreground", "Cold gray daylight mixes with a sickly interior cast, while the melted facial streaks carry a sharp, unnatural crimson-brown accent" ], "visual_inferences": [ "fixed_element cold_gray_interior_lighting: A pervasive cold and gray lighting condition that fills the interior spaces", "No background room described due to skipped close framing; anchored exclusively by the dining table surface", "The subject is a reproduction surface (photograph), so the character ID is replaced with a common noun and demographic descriptor" ], "creative_decisions": [ "Utilized [L07] solely for the table surface to avoid generating background room elements according to the skipped_close_framing rule", "Applied strict reproduction surface rule: using common noun instead of character ID for the photo to prevent full-scale face injection", "Emphasized the crimson-brown streaks against the sickly green-gray ambient light to heighten the supernatural hallucination effect" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "b6389665e9140198", "owned_hash": "f46489d405a4bd51", "camera_direction_hash": "1bf4c911c03fd123", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 18, "dependency_reason": "동일 씬 내 이전 샷(Shot 12)의 차갑고 창백한 실내 조명(cold_gray_interior_lighting) 톤을 연속적으로 유지하되, 클로즈업 프레이밍 정책에 따라 배경 요소는 제외함.", "_shot_index": 15, "visible_entities": [ "P05", "P09", "P10", "L07" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 18, "shot_index": 15 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Subjective POV from 수리영 at dining-table height: a CU, high angle, frozen insert holds on the upright framed photograph near the right-center of the table, with the drinking glass blurred at the left foreground and the frame kept to roughly the central third. The camera inherits the lateral table move but stops dead as the mother's portrait appears to liquefy behind the glass, turning a family snapshot into a hallucinated surface event.", "lighting_mood": "Cold gray daylight mixes with a sickly interior cast, while the melted facial streaks carry a sharp, unnatural crimson-brown accent.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "P05", "P09", "P10", "L07" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_living_kitchen_dawn_photo_hallucination", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "cold_gray_interior_lighting", "element_type": "environment_state", "character_name": "", "description": "A pervasive cold and gray lighting condition that fills the interior spaces. The light casts a pale, sickly, and slightly green-gray tone over the surroundings, creating an eerie, unnatural, and chilling atmosphere across the different rooms.", "applies_to_shots": [ 2, 12, 15 ], "source_facts": [ "Shot 2: 'Thin daylight seeps in cold and gray'", "Shot 12: 'Cold fluorescent bathroom light turns the glass green-gray'", "Shot 15: 'Cold gray daylight mixes with a sickly interior cast'" ], "visual_inferences": [ "A unified desaturated color palette with cold tones to visually connect the different rooms (bedroom, bathroom, dining area) throughout the sequence." ], "creative_decisions": [ "Applying a sickly green-gray tint to the overall ambient light to foreshadow and emphasize the character's psychological instability and the supernatural hallucinations." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 18, "shot_index": 12, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "5fd3aef3cab4cd2f" }, { "scene_index": 19, "heading": "S#19. 마트 / D", "beat_title": "마트 직원과의 대면과 단서 확인", "representative_moment": "수리영(C01)이 종이 박스를 든 마트 직원(C06) 앞을 막아서서 매대 사이에서 눈을 맞추고 있다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Medium two-shot at chest height, squared between shelves, pale blue-gray fluorescent lighting", "t2i_prompt": "Photorealistic cinematic still. A squared medium two-shot at chest height inside the quiet supermarket aisle from the reference, bathed in flat early-morning fluorescent light casting a pale blue-gray wash. C01O02 in a navy windbreaker, a young East Asian woman, occupies the left-middle in three-quarter right profile, blocking the path, her torso and arms in frame. Across the shared aisle space, C06O08 in a red supermarket uniform, a middle-aged East Asian man, occupies the right-middle in three-quarter left profile, his torso in frame, clutching stacked paper boxes high against his chest. The paper boxes form a rigid barrier in the air between them. C01O02 keeps her eyes fixed on C06O08, and C06O08 keeps his eyes fixed on C01O02. Focus on the tense confrontation space marked by the paper boxes wedged between the two figures. The tall shelves from the reference frame them on both sides, firmly anchoring the quiet encounter.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C06", "outlook_id": "O08" } ], "source_facts": [ "S#19 action: 수리영이 어제 만났던 남자 직원을 찾는다", "shot description: 수리영 blocking the path of 남자 직원 holding paper boxes, locking eyes" ], "visual_inferences": [ "employee is carrying stacked boxes as part of morning restocking duties", "chest-height framing fits upper bodies and the clutched boxes naturally into the frame", "supermarket aisle is identical to the prior scene 8 reference" ], "creative_decisions": [ "use the stacked paper boxes as a physical and visual barrier to emphasize tension", "flat pale blue-gray lighting drains warmth to underscore the cold, early-morning atmosphere", "squared two-shot composition visually traps both characters within the parallel lines of the shelves" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "38d5068a00617e52", "owned_hash": "b0062c44273bd902", "camera_direction_hash": "9198f41bcf8f4be0", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 8, "dependency_reason": "Background space binds to the established supermarket layout from S8_Shot5, maintaining exact room details while introducing a new character interaction.", "_shot_index": 3, "visible_entities": [ "C01", "C06", "P05", "L11", "L12" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 19, "shot_index": 3 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At chest height inside the quiet supermarket aisle, the camera ends its parallel track in a MWS, eye-level, squared two-shot framed between tall shelves, holding 수리영 left-middle and 남자 직원 right-middle with the paper boxes wedged between them. The confrontation reads as a still blockage in the aisle: boxed goods form rigid vertical lines behind him while 수리영 cuts across his path without changing the established approach axis.", "lighting_mood": "Flat early-morning supermarket fluorescents wash the aisle in pale blue-gray, draining warmth from the everyday setting.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "C06", "P05", "L11", "L12" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C06", "outlook_id": "O08" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "cb_sales_floor_day_quiet", "reference_usage": "exact_background", "owned_objects": [ "CCTV camera", "cash register", "checkout counter", "display island", "drink case", "emergency exit door", "endcap display", "fluorescent light", "notice board", "refrigerated case", "service shelf", "shelf", "shopping basket", "sliding door", "staff door", "window" ], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "owned_objects lists all environment objects already drawn in the background reference — do not create new ones", "when focusing on an owned object, anchor it explicitly: 'the door from the reference'" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [ { "scene_index": 8, "shot_index": 5, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C06O08", "policy": "required" }, { "kind": "background", "id": "cb_sales_floor_day_quiet", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "0ce6ec40bc35d8d5" }, { "scene_index": 19, "heading": "S#19. 마트 / D", "beat_title": "CCTV 단서 발견", "representative_moment": "A close-up of the round CCTV camera mounted on the supermarket's weather-stained exterior wall.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Close-up, low angle looking up, cold blue-gray overcast tone", "t2i_prompt": "Photorealistic cinematic still. Focus on a round CCTV camera mounted under a dull, damp metal eave, occupying the upper-right of the frame. Low angle looking up. The lens housing is slightly dirty, angled downward, its glass dome catching a small cold highlight. [L12: A weather-stained concrete exterior wall of a commercial building, textured with age and subtle vertical water stains, cuts diagonally behind the camera, extending downward.] Overcast coastal daylight washes the surfaces, casting a muted blue-gray tone over the concrete and metal. No people are visible.", "outfit_assignments": [], "source_facts": [ "Scene text: '마트 밖. CCTV 위치를 확인하다. 찍혔을 만한 위치다.'", "Shot description: 'CU, low angle, isolating the round CCTV under the eave near the upper-right of frame... No people are visible'" ], "visual_inferences": [ "fixed_element: supermarket_exterior_wall with weather-stained concrete, subtle vertical water stains, and muted blue-gray tone", "The damp metal eave provides a top frame border, emphasizing the coastal haze setting" ], "creative_decisions": [ "Low angle pointing up creates a feeling of being watched by a silent, imposing eye", "Placing the CCTV in the upper-right corner leaves the diagonal stained wall to fill the negative space, enhancing the bleak mood" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "672c23126e7150ad", "owned_hash": "59d4c51343060376", "camera_direction_hash": "10e0e3d0b31ad8b6", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 8, "dependency_reason": "Shares the fixed environment state of the supermarket exterior wall established in Scene 8.", "_shot_index": 7, "visible_entities": [ "P05", "L11", "L12" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 19, "shot_index": 7 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Outside at the supermarket corner, the tilt reaches its highest point and tightens into a CU, low angle, isolating the round CCTV under the eave near the upper-right of frame while the corner wall cuts diagonally behind it. No people are visible; the lens housing sits like a mute eye against weather-stained concrete, held as the decisive clue before the camera can drop back toward 수리영.", "lighting_mood": "Overcast coastal daylight keeps the wall muted blue-gray, with the CCTV dome catching a small cold highlight like an accusing eye.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "P05", "L11", "L12" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C06", "outlook_id": "O08" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "cb_exterior_entrance_day_cctv_wall", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "supermarket_exterior_wall", "element_type": "environment_state", "character_name": "", "description": "A weather-stained concrete exterior wall of a commercial building. The surface is textured with age, showing subtle streaks and wear typical of an older structure. It is illuminated by cold, overcast daylight, casting a muted blue-gray tone over the concrete.", "applies_to_shots": [ 7, 8 ], "source_facts": [ "Shot 7: 'weather-stained concrete', 'supermarket corner', 'Overcast coastal daylight keeps the wall muted blue-gray'", "Shot 8: 'Beside the supermarket wall', 'The wall line slides shallowly behind her', 'wall-reflected gray light'" ], "visual_inferences": [ "The concrete has a slightly coarse finish with subtle vertical water stains indicating long-term exposure to coastal weather." ], "creative_decisions": [ "Maintaining a consistent bleak blue-gray color palette on the wall to anchor the spatial continuity between the CCTV reveal and the character's reaction shot." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 8, "shot_index": 6, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C06O08", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "8324985efe2528ec" }, { "scene_index": 19, "heading": "S#19. 마트 / D", "beat_title": "CCTV 위치 파악 및 혜수와의 통화", "representative_moment": "마트 외벽 옆에서 수리영(C01)이 휴대전화(P05)를 귀에 바짝 대고 긴장한 표정으로 통화하는 옆얼굴.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Tight CU, slight Dutch angle, cold overcast daylight with sickly green fluorescent cast", "t2i_prompt": "Photorealistic cinematic still. A slight Dutch angle composition. C01O02 in a navy windbreaker, a young East Asian woman, in right profile, occupying the right half of the frame. Tight framing on her face, neck and upper shoulder visible at the edge. Focus on her tense expression, lips parted in a strained breath, eyes looking into the distance. She is pressing P05 against her ear in the lower-right quadrant, the back edge of the device visible. Illuminated by cold overcast daylight and wall-reflected gray light, with a faint sickly green fluorescent cast catching her jawline. [L12: A weather-stained concrete exterior wall running shallowly behind her, showing subtle vertical streaks and a muted blue-gray tone as a soft background plane.]", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "Shot 8: '휴대폰을 귀에 바짝 댄 채 입을 벌린 수리영의 긴장된 옆얼굴.'", "DP direction: 'tight CU, Dutch angle, profile shot of her tense side face occupying the right half'" ], "visual_inferences": [ "fixed_element supermarket_exterior_wall: weather-stained concrete with a muted blue-gray tone", "The phone (P05) is held in the lower-right quadrant of the frame as she listens intently" ], "creative_decisions": [ "Used 'tight framing on her face, neck and upper shoulder visible' to adhere to the close-framing ID policy and prevent floating head artifacts", "Applied a sickly green fluorescent cast on the jawline to enhance the thriller mood and reflect the light leaking from the storefront" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "d347f63bf296ae66", "owned_hash": "59d4c51343060376", "camera_direction_hash": "63c9f90195e4f103", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 8, "dependency_reason": "동일한 마트 외벽 배경(supermarket_exterior_wall)의 시각적 연속성 유지 (푸른빛이 감도는 낡은 콘크리트 질감).", "_shot_index": 8, "visible_entities": [ "C01", "P05", "L11", "L12" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 19, "shot_index": 8 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Beside the supermarket wall at 수리영's shoulder height, the lateral track resolves into a tight CU, Dutch angle, profile shot of her tense side face occupying the right half while the phone presses against her ear in the lower-right quadrant. The wall line slides shallowly behind her, but the only emphasized change from the CCTV reveal is camera distance, pulled close enough to hold her open mouth and strained breath.", "lighting_mood": "Cold daylight and wall-reflected gray light carve her profile sharply, with a faint sickly green cast from the storefront fluorescents leaking outside.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "P05", "L11", "L12" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C06", "outlook_id": "O08" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "cb_exterior_entrance_day_cctv_wall", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "supermarket_exterior_wall", "element_type": "environment_state", "character_name": "", "description": "A weather-stained concrete exterior wall of a commercial building. The surface is textured with age, showing subtle streaks and wear typical of an older structure. It is illuminated by cold, overcast daylight, casting a muted blue-gray tone over the concrete.", "applies_to_shots": [ 7, 8 ], "source_facts": [ "Shot 7: 'weather-stained concrete', 'supermarket corner', 'Overcast coastal daylight keeps the wall muted blue-gray'", "Shot 8: 'Beside the supermarket wall', 'The wall line slides shallowly behind her', 'wall-reflected gray light'" ], "visual_inferences": [ "The concrete has a slightly coarse finish with subtle vertical water stains indicating long-term exposure to coastal weather." ], "creative_decisions": [ "Maintaining a consistent bleak blue-gray color palette on the wall to anchor the spatial continuity between the CCTV reveal and the character's reaction shot." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 8, "shot_index": 6, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C06O08", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "dab9f21ef60c031b" }, { "scene_index": 20, "heading": "S#20. 마트 총무과 / 실내", "beat_title": "식별 불가능한 CCTV와 수리영의 깊어지는 불안", "representative_moment": "C01 stares anxiously at the monitor, her hands tightly clasped, framed by the blurred backs of the security employee and an unregistered woman.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "MCU, eye-level over-the-shoulder composition, cold bluish monitor glow contrasting with dim blue-gray shadows", "t2i_prompt": "Photorealistic cinematic still. MCU, eye-level over-the-shoulder composition. The blurred shoulder of an East Asian woman and the blurred back of a middle-aged East Asian man occupy the immediate foreground, with their backs to the camera. Below them, the back of a computer monitor with a thin dark bezel and the corner of a cluttered desk form a boxed-in lower frame edge. C01O02 in a navy windbreaker, an East Asian woman, sits centered in the mid-ground, facing the camera. Her hands are tightly clasped near the lower center, and her eyes are fixed downward on the unseen monitor screen. Harsh, cold bluish light from the monitor washes over her anxious face and knuckles, while a sickly fluorescent light pools on the desk surface in the foreground. Focus on C01O02's tense expression. [L13: the space immediately behind her falls into deep, heavy blue-gray shadow, rendering any wall details soft and out of focus].", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "Shot 2: 두 사람 등 뒤에 서서 두 손을 꽉 쥔 채 모니터를 응시하는 수리영의 초조한 정면.", "S#20 action: 수리영 불안이 깊어지고." ], "visual_inferences": [ "fixed_element dim_office_environment: heavy blue-gray shadows, sickly fluorescent light pooling on desk.", "fixed_element cctv_monitor: thin dark bezel, emitting a cold bluish glow that lights C01's face.", "Unregistered characters (혜수, 보안직원) are treated as common nouns with demographic descriptors (East Asian woman, middle-aged East Asian man) in the blurred foreground." ], "creative_decisions": [ "Using the monitor edge and the blurred foreground figures to create a claustrophobic, boxed-in frame around C01 to physically manifest her trapped anxiety.", "Contrasting the harsh cold blue monitor glow on her face against the murky gray-blue background to isolate her emotionally within the scene." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "e2a7a1cd21c7ff67", "owned_hash": "8a3bc4b2c5784cad", "camera_direction_hash": "703b5825d7e91a67", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "N/A", "_shot_index": 2, "visible_entities": [ "C01", "L13" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 20, "shot_index": 2 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From the front corner of the desk at eye level, the camera steadies after its track into an MCU, eye-level, over-the-shoulder composition across the blurred backs of 혜수 and the security employee toward 수리영 centered in the background. Her anxious face sits above her tightly clasped hands near the lower center, with the monitor edge and desk corner forming a boxed-in foreground frame.", "lighting_mood": "Sickly fluorescent light pools on the desk while the room falls into blue-gray office shadow, making 수리영's hands and face feel trapped behind the monitor glow.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "L13" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C02", "outlook_id": "O03" }, { "character_id": "C14", "outlook_id": "O14" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "cb_security_office_day_cctv_review", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "cctv_monitor", "element_type": "persistent_prop", "character_name": "", "description": "A computer monitor resting on a desk, featuring a thin dark bezel. The screen displays extremely dark, blackened digital footage heavily obscured by murky gray-green pixel noise, refresh lines, and compression artifacts. The screen continuously emits a cold, bluish glow.", "applies_to_shots": [ 2, 3 ], "source_facts": [ "scene text: 그 부분만 유독 어두워 식별이 불가능하다.", "Shot 2: the monitor edge and desk corner forming a boxed-in foreground frame... monitor glow", "Shot 3: monitor at just above desk height... blackened footage surrounded by a thin bezel... distorted by pixel noise, refresh lines, and compression darkness... gray-green digital noise." ], "visual_inferences": [ "The monitor is a standard modern flat-screen display used in typical security setups.", "The footage remains statically unidentifiable, showing no clear movement due to the heavy digital noise." ], "creative_decisions": [ "Emphasizing the gray-green noise and the physical thin bezel to strongly anchor the prop as the source of the mystery and frustration in the scene." ], "confidence": "high" }, { "element_id": "dim_office_environment", "element_type": "environment_state", "character_name": "", "description": "The office interior is submerged in heavy, blue-gray shadows, creating a dim and murky atmosphere. The darkness is only broken by a sickly fluorescent light pooling directly on the desk surface and the harsh, cold bluish light cast outward from the computer monitor.", "applies_to_shots": [ 2, 3 ], "source_facts": [ "Shot 2: Sickly fluorescent light pools on the desk while the room falls into blue-gray office shadow", "Shot 3: The monitor casts cold bluish light into the surrounding dim office" ], "visual_inferences": [ "The ambient ceiling lights are mostly turned off or broken, resulting in localized pools of light rather than even illumination." ], "creative_decisions": [ "Using a stark contrast between the sickly fluorescent pooling and the cold blue monitor glow to heighten the sense of anxiety and claustrophobia." ], "confidence": "high" } ], "previous_shot_refs": [], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C02O03", "policy": "required" }, { "kind": "character_outlook", "id": "C14O14", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "5b5e6400d3ade7c0" }, { "scene_index": 20, "heading": "S#20. 마트 총무과 / 실내", "beat_title": "식별 불가한 CCTV", "representative_moment": "어둡고 노이즈 낀 CCTV 화면 속 정체불명의 그림자를 모니터 클로즈업으로 바라본다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Close-up on the monitor screen, slight low angle, cold bluish monitor glow contrasting with murky gray-green pixel noise", "t2i_prompt": "Photorealistic cinematic still. Close-up, slight low angle from just above desk height, positioned directly in front of a computer monitor. The thin dark bezel frames the central screen. The monitor displays extremely dark, blackened digital footage. Within the playback, a vague black shadowy figure stands off-center, heavily obscured by murky gray-green pixel noise, horizontal refresh lines, and thick compression artifacts. Focus on the distorted digital noise and the unidentifiable shadow on the screen surface. The display continuously emits a harsh, cold bluish glow. [L13: At the lower frame edge below the monitor, a dusty desk surface and tangled cables catch a localized pool of sickly fluorescent light, immediately fading into heavy blue-gray shadows].", "outfit_assignments": [], "source_facts": [ "scene text: 그 부분만 유독 어두워 식별이 불가능하다.", "shot description: 검은 그림자만 어렴풋이 보이는 어두운 CCTV 영상이 띄워진 모니터 화면 클로즈업.", "DP direction: directly in front of the monitor at just above desk height... blackened footage surrounded by a thin bezel." ], "visual_inferences": [ "fixed_element cctv_monitor: murky gray-green pixel noise, refresh lines, and compression artifacts.", "fixed_element dim_office_environment: sickly fluorescent light pooling directly on the desk surface.", "The room is entirely excluded due to close framing, leaving only the desk surface and cables visible at the bottom edge." ], "creative_decisions": [ "Framed precisely to isolate the monitor screen as the primary subject, anchoring the bottom frame edge with the desk surface and cables to ground the device physically.", "Used heavy digital artifacting and gray-green noise to prevent the shadowy figure from appearing as a flat, featureless cutout, adding depth through distortion." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "22c60c38f33f33ca", "owned_hash": "0dc9315dd6506f73", "camera_direction_hash": "0cc77d90b6a252ea", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 20, "dependency_reason": "같은 공간(마트 총무과)에서 모니터를 들여다보는 직전 샷(Shot 2)의 환경 상태 및 모니터/조명 상태(fixed_elements)를 연속성 있게 유지", "_shot_index": 3, "visible_entities": [ "L13" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 20, "shot_index": 3 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Directly in front of the monitor at just above desk height, the dolly-in ends in a CU, slight low angle, through-device frame of the CCTV playback, with the screen surface kept central and the blackened footage surrounded by a thin bezel and desk shadows. The vague black shadow in the recorded image sits off-center like a stain, distorted by pixel noise, refresh lines, and compression darkness rather than by the room itself.", "lighting_mood": "The monitor casts cold bluish light into the surrounding dim office, while the CCTV image collapses into murky blacks and gray-green digital noise.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "L13" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C02", "outlook_id": "O03" }, { "character_id": "C14", "outlook_id": "O14" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "cb_security_office_day_dark_cctv_monitor", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "cctv_monitor", "element_type": "persistent_prop", "character_name": "", "description": "A computer monitor resting on a desk, featuring a thin dark bezel. The screen displays extremely dark, blackened digital footage heavily obscured by murky gray-green pixel noise, refresh lines, and compression artifacts. The screen continuously emits a cold, bluish glow.", "applies_to_shots": [ 2, 3 ], "source_facts": [ "scene text: 그 부분만 유독 어두워 식별이 불가능하다.", "Shot 2: the monitor edge and desk corner forming a boxed-in foreground frame... monitor glow", "Shot 3: monitor at just above desk height... blackened footage surrounded by a thin bezel... distorted by pixel noise, refresh lines, and compression darkness... gray-green digital noise." ], "visual_inferences": [ "The monitor is a standard modern flat-screen display used in typical security setups.", "The footage remains statically unidentifiable, showing no clear movement due to the heavy digital noise." ], "creative_decisions": [ "Emphasizing the gray-green noise and the physical thin bezel to strongly anchor the prop as the source of the mystery and frustration in the scene." ], "confidence": "high" }, { "element_id": "dim_office_environment", "element_type": "environment_state", "character_name": "", "description": "The office interior is submerged in heavy, blue-gray shadows, creating a dim and murky atmosphere. The darkness is only broken by a sickly fluorescent light pooling directly on the desk surface and the harsh, cold bluish light cast outward from the computer monitor.", "applies_to_shots": [ 2, 3 ], "source_facts": [ "Shot 2: Sickly fluorescent light pools on the desk while the room falls into blue-gray office shadow", "Shot 3: The monitor casts cold bluish light into the surrounding dim office" ], "visual_inferences": [ "The ambient ceiling lights are mostly turned off or broken, resulting in localized pools of light rather than even illumination." ], "creative_decisions": [ "Using a stark contrast between the sickly fluorescent pooling and the cold blue monitor glow to heighten the sense of anxiety and claustrophobia." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 20, "shot_index": 2, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C02O03", "policy": "required" }, { "kind": "character_outlook", "id": "C14O14", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "a6f86a6bac15826d" }, { "scene_index": 21, "heading": "S#21. 경찰서 안 / D", "beat_title": "끔찍한 증거 사진", "representative_moment": "책상 위에 어지럽게 널린 P11 살인사건 증거 사진 속 끔찍한 상처가 클로즈업된다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "High angle close-up insert from tabletop height, shallow depth of field, harsh cold blue-gray tone.", "t2i_prompt": "Photorealistic cinematic still. High angle close-up insert from tabletop height, shallow depth of field. Focus on the horrific dark red wounds of torn flesh depicted in a glossy P11 printed photograph. The primary photograph lies flat, angled slightly upward toward the camera, overlapping other messy, scattered glossy prints at the margins. The narrow edges of the frame reveal [L15: a scratched office laminate desk surface under cold fluorescent light]. An out-of-focus silver paper clip rests near the frame edge. Harsh overhead fluorescents cast a cold blue-gray wash over the glossy surfaces, allowing the deep crimson of the printed wound to stand out as the only saturated color.", "outfit_assignments": [], "source_facts": [ "S#21 action: '혜수 책상위에는 살인사건 확대 사진들이 어지럽게 널려있다. 목, 어깨, 얼굴등 살점이 떨어져 보는 것만으로도 끔찍하다.'", "shot description: CU insert of a printed crime-scene photograph on the desk, torn wound near center." ], "visual_inferences": [ "fixed_element scattered_crime_photos: chaotic pile of glossy, printed crime-scene photographs scattered and overlapping.", "scratched office laminate desk surface visible only at the extreme edges due to shallow depth of field." ], "creative_decisions": [ "High angle and tabletop height to place the viewer directly over the evidence.", "Cold blue-gray wash contrasting with the deep crimson of the torn flesh to emphasize the gruesome nature of the photo.", "Omitted the wider office background completely per close-framing constraints, using L15 only for the immediate desk surface under the photos." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "3d5e6e7db37919a6", "owned_hash": "aa34738df1198728", "camera_direction_hash": "516d6445544dd91d", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 16, "dependency_reason": "Shares continuous environmental elements (the desk surface) from the established police office space, adhering to close-framing reference constraints.", "_shot_index": 6, "visible_entities": [ "P11", "L15" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 21, "shot_index": 6 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At tabletop height beside 혜수's desk, the camera continues its forward pressure into a CU, high angle, shallow-DOF insert of a printed crime-scene photograph, with the torn wound area held near center and the desk reduced to narrow edges around it. The photograph is not hallucinated but feels invasive: neighboring prints overlap at the margins, making the evidence pile look like something spreading across the desk surface.", "lighting_mood": "Harsh police-office fluorescents flatten the glossy photo in cold blue-gray, with the wound’s dark red tones cutting through as the only saturated color.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "P11", "L15" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C02", "outlook_id": "O03" }, { "character_id": "C04", "outlook_id": "O05" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_office_day_evidence_photos_gps_popup", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "scattered_crime_photos", "element_type": "persistent_prop", "character_name": "", "description": "A chaotic pile of glossy, printed crime-scene photographs scattered and overlapping across a flat office desk. The visible images prominently feature horrific, dark red wounds of torn flesh, serving as a gruesome focal point in the workspace.", "applies_to_shots": [ 6, 7 ], "source_facts": [ "혜수 책상위에는 살인사건 확대 사진들이 어지럽게 널려있다.", "목, 어깨, 얼굴등 살점이 떨어져 보는 것만으로도 끔찍하다." ], "visual_inferences": [ "The photos have a glossy finish that catches the ambient office light.", "They are piled haphazardly, indicating a busy and ongoing investigation." ], "creative_decisions": [ "Positioning the overlapping photographs prominently on the desk to visually connect the insert shot with the character's reaction shot." ], "confidence": "high" }, { "element_id": "cold_police_office_environment", "element_type": "environment_state", "character_name": "", "description": "A sterile police office workspace anchored by a standard, flat bureaucratic desk surface. The surrounding environment is characterized by a cold, desaturated blue-gray tone and institutional stillness, providing a grim backdrop for the evidence and equipment on the desk.", "applies_to_shots": [ 6, 7, 8 ], "source_facts": [ "S#21. 경찰서 안 / D", "사무실에 홀로 앉아있는 수리영" ], "visual_inferences": [ "The desk features a dull, standard institutional surface.", "The overall lighting remains cold and harsh, emphasizing a bleak atmosphere." ], "creative_decisions": [ "Using the edge of the desk and the cold ambient tone as a framing and atmospheric anchor across the close-up shots." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 16, "shot_index": 6, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C02O03", "policy": "required" }, { "kind": "character_outlook", "id": "C04O05", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "e8830f856ec7536c" }, { "scene_index": 21, "heading": "S#21. 경찰서 안 / D", "beat_title": "살인사건 사진을 보고 환각과 일치함에 경악하는 수리영", "representative_moment": "C01 leans over the desk, eyes wide in shock as she looks at the scattered crime-scene photographs.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Dutch angle, seated eye-level looking slightly up, close framing, shallow depth of field with blurred foreground", "t2i_prompt": "Photorealistic cinematic still. A Dutch angle, close framing shot from seated eye level looking slightly upward across a desk. C01O02 in a dark navy windbreaker, a young East Asian woman, leaning forward over the desk edge in a three-quarter left angle facing the camera, her face occupying the upper third of the frame. Focus on her widened, shocked eyes; cold fluorescent light reflects on her pale cheeks and wet eyes, while her neck and shoulders remain visible in the frame. The desk edge forms a horizontal barrier across the lower frame. A chaotic pile of P11 glossy crime-scene photographs lies scattered in the immediate lower foreground on the desk surface, angled toward the camera, their graphic dark red wounds rendered as a soft blur. [L15: The background behind her falls entirely out of focus, showing only desaturated blue-gray office walls and cold institutional overhead lights].", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "혜수 책상위에는 살인사건 확대 사진들이 어지럽게 널려있다.", "넘겨보며 놀라는 수리영, 환각상태에서 본 엄마의 시신모습과 너무도 흡사한 - 목, 어깨, 얼굴등 살점이 떨어져 보는 것만으로도 끔찍하다.", "사무실에 홀로 앉아있는 수리영" ], "visual_inferences": [ "fixed_element scattered_crime_photos on the desk", "cold_police_office_environment providing the desaturated background", "She leans forward over the desk to closely examine the horrific details." ], "creative_decisions": [ "Dutch angle and low-angle tilt upward to visually amplify her psychological shock.", "Placing the photographs in the extreme blurred foreground anchors her reaction to the trigger object without displaying sharp gore.", "Close framing omits the existing background reference entirely, focusing purely on her facial expression and the desk edge." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "be94899be215e3d1", "owned_hash": "aa34738df1198728", "camera_direction_hash": "26b6b6d9dade6188", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 21, "dependency_reason": "Shares the same police office environment and desk setup as the previous shot, though rendered here with close framing and shallow depth of field.", "_shot_index": 7, "visible_entities": [ "C01", "P11", "L15" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 21, "shot_index": 7 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Just across the desk from 수리영 at seated eye level, the upward tilt stops in a CU, Dutch angle, front-side composition with her widened eyes in the upper third and the blurred photographs low in the foreground. She leans into the frame from behind the desk edge, recognizing the echo of her vision while the office behind her falls soft and empty.", "lighting_mood": "Cold fluorescent light catches the wetness in her eyes while the background remains desaturated blue-gray, pushing the shock into a claustrophobic police-office stillness.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "P11", "L15" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C02", "outlook_id": "O03" }, { "character_id": "C04", "outlook_id": "O05" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_office_day_evidence_photos_gps_popup", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "scattered_crime_photos", "element_type": "persistent_prop", "character_name": "", "description": "A chaotic pile of glossy, printed crime-scene photographs scattered and overlapping across a flat office desk. The visible images prominently feature horrific, dark red wounds of torn flesh, serving as a gruesome focal point in the workspace.", "applies_to_shots": [ 6, 7 ], "source_facts": [ "혜수 책상위에는 살인사건 확대 사진들이 어지럽게 널려있다.", "목, 어깨, 얼굴등 살점이 떨어져 보는 것만으로도 끔찍하다." ], "visual_inferences": [ "The photos have a glossy finish that catches the ambient office light.", "They are piled haphazardly, indicating a busy and ongoing investigation." ], "creative_decisions": [ "Positioning the overlapping photographs prominently on the desk to visually connect the insert shot with the character's reaction shot." ], "confidence": "high" }, { "element_id": "cold_police_office_environment", "element_type": "environment_state", "character_name": "", "description": "A sterile police office workspace anchored by a standard, flat bureaucratic desk surface. The surrounding environment is characterized by a cold, desaturated blue-gray tone and institutional stillness, providing a grim backdrop for the evidence and equipment on the desk.", "applies_to_shots": [ 6, 7, 8 ], "source_facts": [ "S#21. 경찰서 안 / D", "사무실에 홀로 앉아있는 수리영" ], "visual_inferences": [ "The desk features a dull, standard institutional surface.", "The overall lighting remains cold and harsh, emphasizing a bleak atmosphere." ], "creative_decisions": [ "Using the edge of the desk and the cold ambient tone as a framing and atmospheric anchor across the close-up shots." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 21, "shot_index": 6, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C02O03", "policy": "required" }, { "kind": "character_outlook", "id": "C04O05", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "bdef60a5660bdfe3" }, { "scene_index": 21, "heading": "S#21. 경찰서 안 / D", "beat_title": "GPS 좌표 확인", "representative_moment": "모니터 화면에 선명하게 떠오른 붉은색 GPS 해상 좌표 데이터.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Eye-level squared close-up on the monitor, cold cyan digital map light punctured by crimson text.", "t2i_prompt": "Photorealistic cinematic still. A glowing computer monitor sits cleanly centered in the frame, its dark bezel clearly visible. The screen displays a digital sea map bathed in cold cyan light, punctured by sharp, glowing crimson red GPS coordinate text positioned slightly below the center. At the lower edge, a sliver of [L15: a flat, sterile bureaucratic desk surface in muted blue-gray darkness] is visible, with a stationary computer mouse resting near the base of the monitor. The cool digital light from the screen faintly reflects on the desk surface beneath it, while the outer frame edges fall into deep shadow. Focus on the glowing screen and the sharp crimson coordinates.", "outfit_assignments": [], "source_facts": [ "S#21 action: '모니터에서 뭔가 뜬다... 엄마 폰 위치추적 결과다. 해상 지도위에 좌표. 37.484400 N / 126. 072486 E'", "shot description: Close-up of a computer monitor showing a sea map and vivid red GPS coordinate text." ], "visual_inferences": [ "fixed_element cold_police_office_environment: A sterile police office workspace anchored by a flat bureaucratic desk surface.", "Cold cyan display light contrasting with deep institutional shadows at the frame edges." ], "creative_decisions": [ "Squared close-up centering the monitor to isolate the digital location reveal.", "Using a sliver of the desk and mouse at the bottom edge to ground the monitor in the physical space without revealing the wider room." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "60bd85988ad887b9", "owned_hash": "aa34738df1198728", "camera_direction_hash": "98dc3757e06b7a77", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 21, "dependency_reason": "같은 경찰서 사무실 책상 환경의 디테일 및 차가운 색감 톤 유지 (이전 샷 참조)", "_shot_index": 8, "visible_entities": [ "P11", "L15" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 21, "shot_index": 8 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At monitor height on the side of the desk, the camera completes its track into a squared CU, eye-level, through-device composition where the display sits cleanly in the center with its bezel and a sliver of desk still visible for scale. The sea map and red GPS coordinate text are readable, the crimson numbers positioned slightly below center like a wound opening in an otherwise bureaucratic interface.", "lighting_mood": "The monitor’s cold cyan map light dominates the frame, punctured by sharp crimson coordinate text against the muted police-office darkness.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "P11", "L15" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C02", "outlook_id": "O03" }, { "character_id": "C04", "outlook_id": "O05" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_office_day_evidence_photos_gps_popup", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "cold_police_office_environment", "element_type": "environment_state", "character_name": "", "description": "A sterile police office workspace anchored by a standard, flat bureaucratic desk surface. The surrounding environment is characterized by a cold, desaturated blue-gray tone and institutional stillness, providing a grim backdrop for the evidence and equipment on the desk.", "applies_to_shots": [ 6, 7, 8 ], "source_facts": [ "S#21. 경찰서 안 / D", "사무실에 홀로 앉아있는 수리영" ], "visual_inferences": [ "The desk features a dull, standard institutional surface.", "The overall lighting remains cold and harsh, emphasizing a bleak atmosphere." ], "creative_decisions": [ "Using the edge of the desk and the cold ambient tone as a framing and atmospheric anchor across the close-up shots." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 21, "shot_index": 6, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C02O03", "policy": "required" }, { "kind": "character_outlook", "id": "C04O05", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "d6d152fa203a8030" }, { "scene_index": 22, "heading": "S#22. 경찰서 밖 / D", "beat_title": "경찰서를 떠나 다급히 자전거를 달리는 수리영", "representative_moment": "수리영(C01)이 자전거 핸들을 꽉 쥐고 도로를 향해 급하게 상체를 숙이며 달리고 있다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "low angle, chest-height side profile action tracking, cold coastal daylight", "t2i_prompt": "Photorealistic cinematic still. Low observer at chest height, upper torso and arms in frame. C01O02 in a navy windbreaker, a young East Asian woman, in left profile, leaning diagonally forward over the handlebars of the bicycle from the reference in a mid-pedal pose. Her hair flows back in the direction opposite to motion. Her clenched hands occupy the lower foreground. Eyes fixed on the open road from the reference stretching across the left. The wet-looking concrete curb from the reference runs parallel along the bottom frame edge, while the muted gray station building from the reference recedes in the background. Cold daylight casts a blue coastal hue over the scene, sharpening the urgency. Focus on her tense profile and her tight grip on the handlebars.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "S#22 action: 주차된 자전거를 타고 이내 달리는 수리영", "shot description: 자전거 핸들을 꽉 쥔 채 도로를 향해 상체를 숙인 수리영의 다급한 측면", "Su-ri-young: profile_left, leaning over handlebars, eyes→road", "lighting mood: Cold daylight over the police-station curb gives the scene a blue coastal cast" ], "visual_inferences": [ "mid-pedal pose and hair flowing backward to convey the mid-action freeze of riding", "chest-height framing captures upper torso and hands while excluding the lower body to emphasize forward momentum" ], "creative_decisions": [ "explicitly anchored the bicycle, road, curb, and station building to 'from the reference' to satisfy the strict background_binding contract", "placed her clenched hands in the lower foreground to serve as a tension anchor within the frame" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "e3c851eeb7c625df", "owned_hash": "4d2f9e2d47e6da8f", "camera_direction_hash": "92c780fae517292b", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "none", "_shot_index": 2, "visible_entities": [ "C01", "L16" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 22, "shot_index": 2 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Parallel to the bicycle on 수리영's side at chest height, the camera tracks with her in an MS, low angle, side-profile action frame, tightening toward the handlebar line and upper torso while the road opens ahead on the left. Her clenched hands occupy the lower foreground without overwhelming the frame, and her bent shoulders drive diagonally forward as she pushes off from the curb into traffic space.", "lighting_mood": "Cold daylight over the police-station curb gives the scene a blue coastal cast, with hard gray asphalt and damp concrete sharpening the urgency.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "L16" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "bg_station_entrance_day_bicycle_departure", "reference_usage": "exact_background", "owned_objects": [ "CCTV pole", "bicycle", "bicycle rack", "bollard", "canopy", "car", "crosswalk", "curb", "parking lot", "parking marking", "plaza", "road", "road arrow", "service door", "sign panel", "sliding door", "station building", "tactile paving", "window" ], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "owned_objects lists all environment objects already drawn in the background reference — do not create new ones", "when focusing on an owned object, anchor it explicitly: 'the door from the reference'" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "background", "id": "bg_station_entrance_day_bicycle_departure", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "457b47441f3f08b1" }, { "scene_index": 23, "heading": "S#23. 어느 PC방 / 실내", "beat_title": "좌표를 확인하는 수리영", "representative_moment": "어두운 PC방 안, 수리영(C01)이 차가운 모니터 불빛을 받으며 지도 좌표를 집중해서 확인하고 있다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Medium close-up from the front-left monitor edge, eye-level, through-object framing with cold blue monitor lighting.", "t2i_prompt": "Photorealistic cinematic still. C01O02 in a navy windbreaker, a young East Asian woman, hunched forward at a three-quarter angle from the right, her eyes fixed intently on the glowing screen. Medium close-up, eye-level, through-object framing. Focus on her intense map-lit expression, her head and upper shoulders occupying the left-center of the frame. The dark plastic bezel of the monitor sits as a blurred sliver in the right foreground. Cold blue-white light from the digital map screen washes across her features and the collar of her windbreaker. The immediate space behind her falls into a shallow depth of field, [L17: dim PC room interior, scuffed plastic cubicle partitions cast in muted gray-black shadow, with a faint sickly fluorescent ambient glow fading into the dark background].", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "S#23 action: '모니터 화면에 앉아있는 수리영, 좌표를 찍고 위치를 확인하고'", "shot description: digital map monitor: on, blue-white map displayed", "pose: three_quarter_right, hunched toward monitor, eyes on monitor", "lighting_mood: Cold blue monitor light, muted gray-black shadow, sickly fluorescent residue overhead" ], "visual_inferences": [ "translated 'carves her face' to 'washes across her features' to prevent the T2I model from generating physical cuts on the character's face (Rule B)", "placed the monitor bezel in the foreground as an out-of-focus element to satisfy the through-object framing requirement" ], "creative_decisions": [ "used shallow depth of field for the close framing to isolate the character's expression and blur the PC room partitions", "anchored the character's neck and upper shoulders in the frame to avoid a disembodied floating face" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "e1ab4178a3198956", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "46c8495429a88c12", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "N/A", "_shot_index": 1, "visible_entities": [ "C01", "P09", "P12", "L17" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 23, "shot_index": 1 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From the flow's front-left monitor edge, use a medium close-up (MCU), eye-level, through-object framing: the monitor bezel sits as a dark sliver in the right foreground while 수리영's map-lit face holds left-center within arm's length. The camera is an observer, not her POV, settling at the end of a quiet dolly-in and holding just long enough for her concentration to read against the indistinct rows of the PC room.", "lighting_mood": "Cold blue monitor light carves her face while the PC방 remains sunk in muted gray-black shadow with sickly fluorescent residue overhead.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "P09", "P12", "L17" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": null, "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "eb107dfa8557e28f" }, { "scene_index": 24, "heading": "S#24. 해안가 / 소나기", "beat_title": "폭우 속을 뚫고 달리는 절박함", "representative_moment": "쏟아지는 폭우 속에서 자전거 핸들을 꽉 쥐고 앞으로 몸을 숙인 채 달리는 C01의 모습.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Front-left tracking camera, eye-level, shallow depth of field, storm-dark blue and slate gray palette with cold highlights.", "t2i_prompt": "Photorealistic cinematic still. Front-left tracking camera at eye-level with a shallow depth of field. C01O02 in a soaked navy windbreaker, a young East Asian woman, occupies the center-left frame, at a three-quarter angle from the left, leaning sharply forward over the bicycle. Her eyes are fixed on the wet road ahead, her wet hair and jacket flap streaming forcefully backward against the motion. Both her clenched hands grip the wet handlebar in the lower third of the frame, the bicycle frame and a slice of the front wheel clearly visible below her arms. Storm-dark blues and slate grays dominate the atmosphere, with brief cold highlights reflecting on the rainwater across her skin and jacket. Heavy wind-driven rain forms a dense atmospheric veil. The rain-slick coastal road and wet roadside guardrail streak behind her as lateral motion blur. Focus on her tense forward lean and clenched hands pushing through the storm.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "S#24 action: '자전거를 타고 빗속을 달리는 수리영. 오로지 엄마를 찾아야한다는 생각 뿐.'", "shot description: 쏟아지는 빗방울 속에서 자전거 핸들을 꽉 쥔 수리영의 젖은 상체, 앞으로 몸을 숙인 채 젖은 머리카락과 옷자락이 등 뒤로 세차게 휘날리며 달리는 고정된 순간" ], "visual_inferences": [ "bicycle frame and front wheel visible below the handlebar to ground the large prop based on system rules", "motion freeze represented by hair and jacket streaming backward against the motion", "background guardrail and road blurred to convey tracking speed" ], "creative_decisions": [ "shallow depth of field to isolate C01's desperate expression and clenched hands against the motion blur of the storm", "cold highlights on wet skin to emphasize the harshness of the weather and the storm-dark mood" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "e6f7e62830fc139d", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "831b1c1357d6a81b", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 9, "dependency_reason": "해안가 도로 배경(S9)의 공간적/시각적 연속성을 유지하되, 폭우라는 날씨 상태의 변형을 적용하여 묘사", "_shot_index": 2, "visible_entities": [ "C01", "L08" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 24, "shot_index": 2 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Maintain the flow's front-left tracking position with a medium shot (MS), eye-level, shallow DOF frame held unusually steady inside the motion: 수리영's wet upper body occupies center-left, both clenched hands visible near the lower third on the handlebar. The camera runs just ahead of the bicycle, compressing rain, wind-whipped hair, and forward lean into a single forceful instant while the slick coastal road streaks behind her.", "lighting_mood": "Storm-dark blues and slate grays are cut by brief cold highlights on rainwater, making her skin and soaked clothes look harsh and desperate.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "L08" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "not_applicable", "bg_id": null, "reference_usage": "none", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "no chain_bg is bound to this shot — describe the location freely without referencing a background image", "do not use 'from the reference' / 'the existing room' / 'preserving the room' style phrasing — there is no reference image to refer to" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [ { "scene_index": 9, "shot_index": 2, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "8e3add8d9656925b" }, { "scene_index": 25, "heading": "S#25. 옥탑방 안 / 해질 무렵", "beat_title": "침입의 흔적 발견", "representative_moment": "바닥에 남겨진 진흙 묻은 낯선 발자국들을 발견한다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Worm's eye view at floor level with shallow depth of field, cool gray and greenish shadows", "t2i_prompt": "Photorealistic cinematic still. A view from almost floor level at the entrance threshold. Fresh, wet, dark muddy footprints dominate the sharp foreground, running diagonally from the lower left across [L07: a pale beige vinyl floor of a dim living room]. Focus on the thick mud textures. Fading cold gray dusk light leaks across the floor in sharp bands, casting heavy dull shadows with a faint greenish cast, turning the mud nearly black against the sickly beige vinyl. In the soft, out-of-focus background, overturned dining chairs and a small open window are visible in the distance, admitting the gloomy outside air. Tense, eerie atmosphere without any visible figures.", "outfit_assignments": [], "source_facts": [ "S#25 action: '누군가 다녀간 흔적으로 보이는 발자국들. 집안에서 뭔가를 찾기 위해 가재도구를 온통 헤집은 것으로 보인다. 거실에 난 창은 열려있고, 식탁 의자들을 쓰러져있다.'", "shot description: 거실 장판 바닥 곳곳에 선명하게 찍힌 진흙 묻은 낯선 발자국들의 클로즈업" ], "visual_inferences": [ "fixed_element cold_dusk_lighting: fading dusk light with cold gray bands and greenish cast", "fixed_element ransacked_living_room: overturned furniture and open window in the distant background", "Observer POV is depicted as a direct floor-level view of the prints, leaving characters out of the frame" ], "creative_decisions": [ "Floor-level camera placement to emphasize the texture and direction of the muddy footprints", "Soft, blurred background framing to comply with the close-framing scale while maintaining continuity of the ransacked room" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "0d3818211eab29f4", "owned_hash": "80f679b464a34fb0", "camera_direction_hash": "f4dafde5c706768e", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 12, "dependency_reason": "동일한 공간 및 상황(이전 샷 또는 플래시백 12씬 연관)과의 연속성 유지", "_shot_index": 2, "visible_entities": [ "P08", "P09", "L07" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 25, "shot_index": 2 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Drop below the flow default into an extreme close-up (ECU), worm's eye angle, almost at floor level from the entrance threshold, looking inward across the 장판. Muddy unfamiliar footprints run diagonally from the lower left toward the dim living room, sharp in the foreground while overturned furniture and the open window remain small, blurred scale references in the distance.", "lighting_mood": "Fading dusk light leaks across the floor in cold gray bands, turning the mud nearly black against the sickly beige vinyl.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "P08", "P09", "L07" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_living_kitchen_dusk_ransacked_windy", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "cold_dusk_lighting", "element_type": "environment_state", "character_name": "", "description": "The interior space is illuminated by fading, cold gray dusk light, mixed with heavy dull shadows and a faint greenish cast, creating a gloomy and suspenseful atmosphere.", "applies_to_shots": [ 2, 5, 9 ], "source_facts": [ "Shot 2: Fading dusk light leaks across the floor in cold gray bands", "Shot 5: Dim sunset gray mixes with a faint interior green cast", "Shot 9: Cold dusk light from the open window and dull interior shadow" ], "visual_inferences": [ "The time of day creates a consistent cold, low-light environment across both the living room and the inner bedroom." ], "creative_decisions": [ "Unified the overall color palette with cold grays and faint greens to emphasize the eerie occult theme of the scene." ], "confidence": "high" }, { "element_id": "ransacked_living_room", "element_type": "environment_state", "character_name": "", "description": "A chaotic living room area featuring a pale beige vinyl floor stamped with dark, muddy footprints. In the background, pieces of furniture are overturned and scattered, and a window is left wide open, letting in the outside air.", "applies_to_shots": [ 2, 9 ], "source_facts": [ "Shot 2: Muddy unfamiliar footprints... sickly beige vinyl... overturned furniture and the open window remain small", "Shot 9: Cold dusk light from the open window... dollies out along the entrance path" ], "visual_inferences": [ "The muddy footprints and overturned furniture established in Shot 2 remain in the background and on the floor of Shot 9 as the character walks through the same living room entrance path." ], "creative_decisions": [ "Kept the open window and muddy footprints visually present in the living room framing to ensure visual continuity of the recent break-in across both angles." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 12, "shot_index": 20, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "6c4288d3e13a4f1b" }, { "scene_index": 25, "heading": "S#25. 옥탑방 안 / 해질 무렵", "beat_title": "붉은 원과 피묻은 사진의 발견", "representative_moment": "벽지 한가운데 그려진 커다란 붉은 원과 그 정중앙에 꽂힌 과거의 사진이 스산한 방 안의 중심을 차지한다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "eye-level medium framing, straight-on view, cold gray dusk lighting with a faint green cast", "t2i_prompt": "Photorealistic cinematic still. Focus on the blood-stained photo from the reference, pinned to the exact center of the large red circle from the reference on the aged wallpaper from the reference. The interior space is illuminated by fading, cold gray dusk light mixed with heavy dull shadows and a faint greenish cast. The shadowed door from the reference and the rumpled edge of the bed from the reference sit at the frame margins, framing the central graphic threat. The crimson circle stands out as the only aggressive color against the gloomy, low-light environment.", "outfit_assignments": [], "source_facts": [ "Shot 5: 벽지 한가운데 붉은색으로 칠해진 커다란 원형 표식과 그 정중앙에 꽂힌 사진 한 장.", "DP notes: graphic reveal of the marked wallpaper. The red circle sits centered on the back wall with the blood-stained photograph pinned at its exact middle", "DP notes: bed edge and open doorway stay at the frame margins so the symbol reads like an occult target without losing the cramped room's realism." ], "visual_inferences": [ "fixed_element cold_dusk_lighting: The interior space is illuminated by fading, cold gray dusk light, mixed with heavy dull shadows and a faint greenish cast.", "Character removed from frame per user request; the shot serves as a pure POV or environmental focus on the wall markings." ], "creative_decisions": [ "Used medium framing to ensure both the central photo/circle and the peripheral room edges (door, bed) are visible, anchoring the realism of the cramped room.", "Explicitly appended 'from the reference' to all owned objects (photo, circle, wallpaper, door, bed) as required by the background_ref_attached constraint." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "3709018a66fc4a84", "owned_hash": "36aeda8df398abde", "camera_direction_hash": "d6ec4c2026a8b3b1", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "photo", "violating_phrase": "Focus on the blood-stained photo from the reference, pinned to the exact center", "reason": "It describes and effectively reintroduces the owned photo as a newly staged focal object rather than only referencing the existing one." }, { "owned_object": "circle", "violating_phrase": "the exact center of the large red circle from the reference", "reason": "It directs the circle to be rendered as an explicit graphic element, which reads as redrawing an owned background object." }, { "owned_object": "wallpaper", "violating_phrase": "on the aged wallpaper from the reference", "reason": "It uses the wallpaper as a specifically re-created surface element with added emphasis, implying a redraw of the owned background wallpaper." }, { "owned_object": "door", "violating_phrase": "The shadowed door from the reference", "reason": "It refers to the door as a visible subject in the new composition, which can imply re-creating the owned door rather than only anchoring to it." }, { "owned_object": "bed", "violating_phrase": "the rumpled edge of the bed from the reference", "reason": "It explicitly places the bed edge as part of the staged shot, which implies re-drawing the owned bed." } ] } } ], "scene_type": "normal", "dependent_scene_index": 25, "dependency_reason": "Same room, exact background carried over from the prior shots in the sequence.", "_shot_index": 5, "visible_entities": [ "P08", "P09", "L07" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 25, "shot_index": 5 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From just inside 수리영's room, keep the flow's straight-on push and resolve into a wide shot (WS), eye-level, graphic reveal of the marked wallpaper. The red circle sits centered on the back wall with the blood-stained photograph pinned at its exact middle, while the bed edge and open doorway stay at the frame margins so the symbol reads like an occult target without losing the cramped room's realism.", "lighting_mood": "Dim sunset gray mixes with a faint interior green cast, making the crimson circle the only aggressive color in the room.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "P08", "P09", "L07" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "bg_daughter_bedroom_dusk_red_circle_bloody_photo", "reference_usage": "exact_background", "owned_objects": [ "backpack", "bed", "blanket", "ceiling light", "circle", "curtain", "doll", "door", "floor", "photo", "wall", "wallpaper", "window" ], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "owned_objects lists all environment objects already drawn in the background reference — do not create new ones", "when focusing on an owned object, anchor it explicitly: 'the door from the reference'" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "cold_dusk_lighting", "element_type": "environment_state", "character_name": "", "description": "The interior space is illuminated by fading, cold gray dusk light, mixed with heavy dull shadows and a faint greenish cast, creating a gloomy and suspenseful atmosphere.", "applies_to_shots": [ 2, 5, 9 ], "source_facts": [ "Shot 2: Fading dusk light leaks across the floor in cold gray bands", "Shot 5: Dim sunset gray mixes with a faint interior green cast", "Shot 9: Cold dusk light from the open window and dull interior shadow" ], "visual_inferences": [ "The time of day creates a consistent cold, low-light environment across both the living room and the inner bedroom." ], "creative_decisions": [ "Unified the overall color palette with cold grays and faint greens to emphasize the eerie occult theme of the scene." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 25, "shot_index": 2, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "background", "id": "bg_daughter_bedroom_dusk_red_circle_bloody_photo", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "20546db5d6786df3", "status": "contract_violation" }, { "scene_index": 25, "heading": "S#25. 옥탑방 안 / 해질 무렵", "beat_title": "결연한 발걸음", "representative_moment": "C01이 손에 낡은 사진을 꽉 쥔 채 굳은 표정으로 카메라를 향해 걸어온다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Medium close-up, low angle from chest height, cold dusk lighting with a faint greenish cast", "t2i_prompt": "Photorealistic cinematic still. C01O02 in a dark windbreaker, a young Asian woman, advancing toward the camera in mid-step. Focus on her resolute frontal stare and the tightly gripped photograph at the lower frame edge. Shot from a low angle at chest height, her head and upper torso occupy the center, with the ceiling and upper walls extending behind her. She is facing the camera, her eyes fixed directly forward at the camera. The thick straps of P09, a backpack, pull tightly across both her shoulders. One hand firmly grips P08, a small photograph marked with dark red stains, held near the lower right of her torso just entering the bottom frame. The scene is illuminated by fading cold gray dusk light flattening her features, mixed with a sickly faint green interior shadow. [L07: heavily blurred frame-edge walls and indistinct out-of-focus shapes of overturned household items, framing her narrow path in deep dimness].", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "S#25 action: 수리영, 마치 이끌리듯 백팩을 메고 밖으로 나간다. 사진을 손에 꼭 쥔 채.", "shot description: 수리영 advancing into the center while the camera dollies out... backpack straps over both shoulders... clenched photograph near lower right.", "DP directive: Hold the flow's frontal retreat as a medium close-up (MCU), low angle from chest height." ], "visual_inferences": [ "fixed_element cold_dusk_lighting: fading cold gray dusk light mixed with a faint greenish cast.", "fixed_element ransacked_living_room: blurred shapes of overturned furniture kept in the background for visual continuity.", "The forward motion is captured as a mid-step freeze with her upper body leaning slightly into the movement." ], "creative_decisions": [ "Positioned the camera at a low angle from chest height to make her advance feel determined and imposing.", "Replaced the word 'cut' with 'pull tightly' for the backpack straps to avoid the T2I model misinterpreting it as physical bodily harm.", "Blurred the background heavily to comply with close framing background skipped rules while still hinting at the ransacked state." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "38dd212e3a982366", "owned_hash": "80f679b464a34fb0", "camera_direction_hash": "94232b33af724f11", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 25, "dependency_reason": "Shot 5 established the overall interior lighting and the state of the ransacked living room, which is carried over as ambient mood and blurred background elements here.", "_shot_index": 9, "visible_entities": [ "C01", "P08", "P09", "L07" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 25, "shot_index": 9 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Hold the flow's frontal retreat as a medium close-up (MCU), low angle from chest height, with 수리영 advancing into the center while the camera dollies out along the entrance path. Her backpack straps cut dark lines over both shoulders, and the clenched photograph stays visible near the lower right of her torso, making the frontal stare feel like a decision rather than a neutral portrait.", "lighting_mood": "Cold dusk light from the open window and dull interior shadow flatten the room, while the photo's dark red stains echo the occult mark she has just left behind.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "P08", "P09", "L07" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_living_kitchen_dusk_ransacked_windy", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "cold_dusk_lighting", "element_type": "environment_state", "character_name": "", "description": "The interior space is illuminated by fading, cold gray dusk light, mixed with heavy dull shadows and a faint greenish cast, creating a gloomy and suspenseful atmosphere.", "applies_to_shots": [ 2, 5, 9 ], "source_facts": [ "Shot 2: Fading dusk light leaks across the floor in cold gray bands", "Shot 5: Dim sunset gray mixes with a faint interior green cast", "Shot 9: Cold dusk light from the open window and dull interior shadow" ], "visual_inferences": [ "The time of day creates a consistent cold, low-light environment across both the living room and the inner bedroom." ], "creative_decisions": [ "Unified the overall color palette with cold grays and faint greens to emphasize the eerie occult theme of the scene." ], "confidence": "high" }, { "element_id": "ransacked_living_room", "element_type": "environment_state", "character_name": "", "description": "A chaotic living room area featuring a pale beige vinyl floor stamped with dark, muddy footprints. In the background, pieces of furniture are overturned and scattered, and a window is left wide open, letting in the outside air.", "applies_to_shots": [ 2, 9 ], "source_facts": [ "Shot 2: Muddy unfamiliar footprints... sickly beige vinyl... overturned furniture and the open window remain small", "Shot 9: Cold dusk light from the open window... dollies out along the entrance path" ], "visual_inferences": [ "The muddy footprints and overturned furniture established in Shot 2 remain in the background and on the floor of Shot 9 as the character walks through the same living room entrance path." ], "creative_decisions": [ "Kept the open window and muddy footprints visually present in the living room framing to ensure visual continuity of the recent break-in across both angles." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 25, "shot_index": 5, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "5eef56dd826fa350" }, { "scene_index": 26, "heading": "S#26. 포구 / 해질 무렵", "beat_title": "안개 속에서 다가오는 어선", "representative_moment": "짙은 안개벽을 뚫고 낡은 어선이 검은 형체를 드러내며 포구 선착장으로 다가온다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "low angle wide composition from the pier edge, muted cold blue-gray lighting", "t2i_prompt": "Photorealistic cinematic still. Focus on the dark bow of an old small fishing boat emerging from a thick gray fog bank. Low angle observer view from the pier edge, with wet, slick wooden dock boards occupying the lower foreground, receding toward the dark water. The boat is positioned center-frame, cutting through the rippled water directly toward the camera. Its hull appears as a heavy dark shape against the pale fog, its volume preserved by a faint subtle sheen on the damp wood and dim water reflections on the waterline. A pervasive, cold blue-gray fog blankets the background, swallowing the rear half of the boat and severely limiting visibility. The ambient lighting is a muted, bleak coastal haze, desaturating the environment into deep grays and blues.", "outfit_assignments": [], "source_facts": [ "shot description: old small fishing boat's bow appears center-frame as a half-revealed black silhouette cutting the water", "shot description: wet dock boards in the lower foreground", "fixed_element: A thick, pervasive gray fog blankets the coastal harbor and the sea, severely limiting background visibility", "fixed_element: The ambient lighting is a muted, cold blue-gray, creating a damp atmosphere" ], "visual_inferences": [ "low camera at the pier places the wet dock boards prominently in the lower frame, anchoring the viewer's position", "the boat is not a completely flat black cutout, but retains faint volume and moisture sheen to stay photorealistic" ], "creative_decisions": [ "using the contrast between the dark, textured wet foreground boards and the pale, flat background fog to emphasize depth", "framing the boat strictly in the center to highlight its omen-like arrival through the haze" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "509b0707b2e281d0", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "bdd4ae438ee3f2e8", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "", "_shot_index": 3, "visible_entities": [ "P05", "P13", "L18", "L19" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 26, "shot_index": 3 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At the pier edge, use a wide shot (WS), low angle with a slightly long lens, aimed straight into the fog as established by the flow. The old fishing boat's bow appears center-frame as a half-revealed black silhouette cutting the water, with wet dock boards in the lower foreground and most of the hull swallowed by the gray fog bank.", "lighting_mood": "A cold coastal haze desaturates the harbor into blue-gray layers, leaving the boat as a nearly black omen against pale fog.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "P05", "P13", "L18", "L19" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C03", "outlook_id": "O04" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "not_applicable", "bg_id": null, "reference_usage": "none", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "no chain_bg is bound to this shot — describe the location freely without referencing a background image", "do not use 'from the reference' / 'the existing room' / 'preserving the room' style phrasing — there is no reference image to refer to" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "dense_harbor_fog", "element_type": "environment_state", "character_name": "", "description": "A thick, pervasive gray fog blankets the coastal harbor and the sea, severely limiting background visibility. The ambient lighting is a muted, cold blue-gray, creating a damp atmosphere that softens distant shapes and desaturates the overall environment.", "applies_to_shots": [ 3, 6, 9 ], "source_facts": [ "Scene text states the rain stopped and thick fog rolled in (비는 그치고 안개가 자욱하다)", "Shot 3 explicitly mentions 'gray fog bank' and 'blue-gray layers'", "Shot 6 references 'Flat blue-gray fog light softens the background'", "Shot 9 notes 'foggy water' and 'Muted coastal blue-gray light'" ], "visual_inferences": [ "The fog acts as a natural light diffuser, removing harsh shadows and rendering the illumination flat and gloomy", "Environmental surfaces such as the dock boards appear damp or wet due to the recent rain and heavy fog" ], "creative_decisions": [ "Using the thick fog to visually isolate the setting from the rest of the world, emphasizing the bleak, lonely mood of the isolated harbor" ], "confidence": "high" } ], "previous_shot_refs": [], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C03O04", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "2d39136d819f90c9" }, { "scene_index": 26, "heading": "S#26. 포구 / 해질 무렵", "beat_title": "안개 낀 포구에서 배를 찾은 수리영과 무심한 인우", "representative_moment": "안개 속에서 밧줄을 쥔 채 다가오는 수리영(C01)을 힐끗 쳐다보는 인우(C03)의 클로즈업.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Eye-level close-up with flat blue-gray fog lighting and shallow depth of field", "t2i_prompt": "Photorealistic cinematic still. Eye-level close-up composition. The primary subject is C03O04 in a green waterproof jacket, a young Asian man, at three-quarter angle from the right, wearing an indifferent expression. Tight framing on his face, upper body and shoulder in frame. He loosely grips a thick, wet mooring rope visible along the lower frame edge. His eyes are fixed across the hazy air on C01O02 in a navy windbreaker, a young Asian woman, at three-quarter angle from the left. She is rendered as a soft, out-of-focus background figure, mid-stride toward the foreground with one foot slightly lifted off the ground, her eyes fixed back on him. The two figures share the same wet pier they both occupy. Flat blue-gray fog light softens the environment, casting damp highlights on C03O04's cheek and the textured rope. A thick, pervasive gray fog blankets the coastal background behind them, severely limiting visibility.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C03", "outlook_id": "O04" } ], "source_facts": [ "Scene text describes a thick fog blanketing the empty harbor after the rain.", "Shot description specifies In-woo holding a mooring rope and glancing indifferently at Su-ri-young.", "Su-ri-young approaches the boat while In-woo secures the line." ], "visual_inferences": [ "fixed_element dense_harbor_fog: pervasive gray fog acting as a natural light diffuser.", "Water droplets and damp highlights on the rope and In-woo's face due to the recent rain and heavy fog.", "Su-ri-young is kept out of focus in the background to emphasize In-woo's reaction in the close-up." ], "creative_decisions": [ "Used 'tight framing on his face, upper body and shoulder in frame' to comply with close-framing constraints for faces.", "Included the shared anchor 'the same wet pier they both occupy' to logically connect the foreground and background figures in the space.", "Specified 'mid-stride toward the foreground with one foot slightly lifted' to naturally freeze Su-ri-young's approaching motion." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "43934b6c46a5f094", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "146e8f1f047d2725", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 26, "dependency_reason": "Same scene continuous progression, retaining the established thick fog and blue-gray lighting conditions.", "_shot_index": 6, "visible_entities": [ "C01", "C03", "P05", "P13", "L18", "L19" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 26, "shot_index": 6 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From close beside the boat, tighten the flow into a close-up (CU), eye-level, slight over-the-shoulder geometry from just forward of 인우's shoulder back toward his face. The rope remains implied along the lower edge while 인우's indifferent glance lands across frame toward 수리영, who is held smaller in the pier background along his eyeline.", "lighting_mood": "Flat blue-gray fog light softens the background while damp highlights on rope and skin keep the close-up gritty and restrained.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "C03", "P05", "P13", "L18", "L19" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C03", "outlook_id": "O04" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": null, "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "dense_harbor_fog", "element_type": "environment_state", "character_name": "", "description": "A thick, pervasive gray fog blankets the coastal harbor and the sea, severely limiting background visibility. The ambient lighting is a muted, cold blue-gray, creating a damp atmosphere that softens distant shapes and desaturates the overall environment.", "applies_to_shots": [ 3, 6, 9 ], "source_facts": [ "Scene text states the rain stopped and thick fog rolled in (비는 그치고 안개가 자욱하다)", "Shot 3 explicitly mentions 'gray fog bank' and 'blue-gray layers'", "Shot 6 references 'Flat blue-gray fog light softens the background'", "Shot 9 notes 'foggy water' and 'Muted coastal blue-gray light'" ], "visual_inferences": [ "The fog acts as a natural light diffuser, removing harsh shadows and rendering the illumination flat and gloomy", "Environmental surfaces such as the dock boards appear damp or wet due to the recent rain and heavy fog" ], "creative_decisions": [ "Using the thick fog to visually isolate the setting from the rest of the world, emphasizing the bleak, lonely mood of the isolated harbor" ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 26, "shot_index": 3, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C03O04", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "5a9866959b85a446" }, { "scene_index": 26, "heading": "S#26. 포구 / 해질 무렵", "beat_title": "안개 속 바다를 향해 남기는 마지막 음성메시지", "representative_moment": "배에 오르기 전, 수리영(C01)이 짙은 안개가 깔린 바다를 바라보며 핸드폰에 음성메시지를 남긴다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "near-static medium close-up (MCU), slight high angle, muted coastal blue-gray lighting", "t2i_prompt": "Photorealistic cinematic still. Head-and-shoulder composition from a slight high angle. C01O02 in a navy windbreaker, a young East Asian woman, occupies the left third of the frame in right profile, the top of her head and upper chest in frame. She holds P05, a smartphone, near her lips, the screen turned away from the camera. Her eyes are fixed on the distant sea. Muted coastal blue-gray light wraps her profile in damp softness, and a faint cold highlight catches on the phone and her lips. [L18: a wet, dark pier railing edge occupies the lower right frame, the background completely swallowed by a thick, pervasive gray fog, desaturating the space into a cold blue-gray haze]. Focus on her side profile and the phone near her mouth.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "Scene text states the rain stopped and thick fog rolled in (비는 그치고 안개가 자욱하다)", "수리영 leaves a voice message looking at the sea before departing (수리영, 떠나기전, 바다를 바라보며 음성메시지를 남긴다)", "Observer perspective, MCU, slight high angle, clean side profile." ], "visual_inferences": [ "fixed_element dense_harbor_fog: A thick, pervasive gray fog blankets the coastal harbor, muting the lighting to a cold blue-gray.", "The wet pier railing edge anchors the lower frame, visually grounding her near the water while the fog isolates her.", "The phone held near her lips catches a faint cold highlight from the ambient fog light." ], "creative_decisions": [ "Placing C01 in the left third facing right leaves empty space for the foggy sea, emphasizing her sense of departure and isolation.", "Applying a slight high angle head-and-shoulder composition keeps her connected to the pier environment while avoiding a floating facial close-up." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "d47869b25e76a069", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "650b1d399555aeb5", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 26, "dependency_reason": "Shot 3 established the thick gray fog, damp coastal atmosphere, and muted blue-gray lighting that defines this sequence.", "_shot_index": 9, "visible_entities": [ "C01", "P05", "P13", "L18", "L19" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 26, "shot_index": 9 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Settle the flow's seaward-side track into a near-static medium close-up (MCU), slight high angle, profile framing across 수리영's cheek line toward the foggy water. 수리영 occupies the left third in clean side profile with the phone at her mouth, while the waiting boat and gray sea sit beyond her gaze rather than showing her boarding yet.", "lighting_mood": "Muted coastal blue-gray light wraps her profile in damp softness, with the phone and lips catching the faintest cold highlight.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "P05", "P13", "L18", "L19" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C03", "outlook_id": "O04" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": null, "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "dense_harbor_fog", "element_type": "environment_state", "character_name": "", "description": "A thick, pervasive gray fog blankets the coastal harbor and the sea, severely limiting background visibility. The ambient lighting is a muted, cold blue-gray, creating a damp atmosphere that softens distant shapes and desaturates the overall environment.", "applies_to_shots": [ 3, 6, 9 ], "source_facts": [ "Scene text states the rain stopped and thick fog rolled in (비는 그치고 안개가 자욱하다)", "Shot 3 explicitly mentions 'gray fog bank' and 'blue-gray layers'", "Shot 6 references 'Flat blue-gray fog light softens the background'", "Shot 9 notes 'foggy water' and 'Muted coastal blue-gray light'" ], "visual_inferences": [ "The fog acts as a natural light diffuser, removing harsh shadows and rendering the illumination flat and gloomy", "Environmental surfaces such as the dock boards appear damp or wet due to the recent rain and heavy fog" ], "creative_decisions": [ "Using the thick fog to visually isolate the setting from the rest of the world, emphasizing the bleak, lonely mood of the isolated harbor" ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 26, "shot_index": 3, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C03O04", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "6ffc5ee05fbc125d" }, { "scene_index": 27, "heading": "S#27. 옥탑방 안 / 해질 무렵", "beat_title": "수리영의 메시지를 듣고 불안해하는 혜수", "representative_moment": "C02 stands in the middle of the dim room, looking down anxiously with the phone pressed to her ear.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Eye-level medium close-up, sickly green-gray ambient mixed with fading dusk light", "t2i_prompt": "Photorealistic cinematic still. Eye-level medium close-up, head-and-shoulder composition. C02O03 in a gray leather jacket, a young East Asian woman, at three-quarter angle from the left, looking down with an anxious, serious expression. One hand holds P05, a mobile phone, with its back and edge visible pressed tightly against her ear. Sickly interior green-gray light washes across her face, flattening her features, mixed with cool fading dusk ambient pooling on her shoulder. Focus on her tense expression and the phone at her ear. [L07: Immediately behind her, deep shadows and out-of-focus scattered debris form a cold, dim, green-gray background of the ransacked space].", "outfit_assignments": [ { "character_id": "C02", "outlook_id": "O03" } ], "source_facts": [ "S#27 action: 혜수 is checking a voice message, realizing something is wrong.", "Shot 1 description: 혜수 stands in the middle of the ransacked rooftop room, phone pressed to her ear, serious face.", "DP direction: MCU, eye-level, tightened so her face and phone dominate center frame." ], "visual_inferences": [ "fixed_element 'ransacked_rooftop_interior' adapted for close framing as soft background debris and dim gloom behind her.", "The sickly green-gray dusk light flattens the lighting on her face, emphasizing a cold, anxious mood.", "The back and edge of the phone are visible against her ear as instructed." ], "creative_decisions": [ "Used a head-and-shoulder composition to satisfy the 'close' framing scale without triggering face-only detachment issues.", "Applied 'skipped_close_framing' rules by describing only the immediate frame-edge background without invoking the exact prior room reference.", "Restricted focus entirely to C02's expression and the phone, omitting other characters from this intimate observer shot." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "b42ed3bd730a67d2", "owned_hash": "80f679b464a34fb0", "camera_direction_hash": "32d2be677a626134", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 12, "dependency_reason": "Prior S12 shot establishes the location, but this shot uses close framing which skips the exact background reference while maintaining the color and mood.", "_shot_index": 1, "visible_entities": [ "C02", "P05", "L04", "L07" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 27, "shot_index": 1 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Begin the scene flow with a medium close-up (MCU), eye-level, intimate observer framing about two meters in front of 혜수, tightened so her face and phone dominate center frame. The ransacked rooftop room stays visible behind her in broken layers, but the composition holds on her anxious listening before the camera will later slide away toward the entrance.", "lighting_mood": "Sickly interior green-gray mixes with dying dusk, flattening 혜수's face while the room behind her feels emptied and cold.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C02", "P05", "L04", "L07" ], "allowed_outlook_pairs": [ { "character_id": "C02", "outlook_id": "O03" }, { "character_id": "C04", "outlook_id": "O05" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_living_kitchen_dusk_ransacked_windy", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "ransacked_rooftop_interior", "element_type": "environment_state", "character_name": "", "description": "The interior of an old rooftop room is left in a messy, ransacked state with scattered objects and debris lying around in broken layers. The entire space is heavily shadowed and enveloped in a cold, sickly green-gray gloom, dimly illuminated by the fading light of dusk.", "applies_to_shots": [ 1, 4 ], "source_facts": [ "Shot 1: The ransacked rooftop room stays visible behind her in broken layers", "Shot 1: Sickly interior green-gray mixes with dying dusk", "Shot 4: gray-green interior gloom", "Shot 4: cold dusk light leaks through" ], "visual_inferences": [ "The chaotic placement of objects from being ransacked remains untouched across the scene.", "The ambient light level is consistently dim, relying entirely on the minimal dusk light leaking from outside." ], "creative_decisions": [ "Maintain a consistent green-gray color grading to emphasize the bleak, empty, and anxious atmosphere of the scene, visually tying the isolated frames together." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 12, "shot_index": 20, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C02O03", "policy": "required" }, { "kind": "character_outlook", "id": "C04O05", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "4ca22c07a7ffe698" }, { "scene_index": 27, "heading": "S#27. 옥탑방 안 / 해질 무렵", "beat_title": "불안감을 증폭시키는 덜컹이는 낡은 현관문", "representative_moment": "바람에 밀려 살짝 틈이 벌어진 낡은 현관 철문 사이로 차가운 석양빛이 스며든다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Close-up, low angle through-gap composition with cold dusk light leaking in", "t2i_prompt": "Photorealistic cinematic still. Close-up on an old, scuffed metal entrance door pushed slightly open. Viewed from a low angle, the dusty concrete threshold occupies the lower frame. Focus on the narrow, off-center gap between the heavy door and its frame. A thin strip of exterior cold gray light leaks through the opening, illuminating the worn metal edge. The surrounding frame edges reveal [L07: a bare interior wall enveloped in a heavily shadowed, sickly green-gray gloom], creating a tense and desolate atmosphere.", "outfit_assignments": [], "source_facts": [ "S#27 action: 덜커덩 덜커덩. 바람에 현관문 소리. 황량하다.", "shot description: 강한 바람에 밀려 살짝 틈이 벌어진 낡은 현관 철문." ], "visual_inferences": [ "fixed_element interior state: cold, sickly green-gray gloom dimly illuminated by fading dusk light", "low angle camera necessitates the threshold/floor being visible at the bottom of the frame" ], "creative_decisions": [ "Focus purely on the door gap and the leaking light as a standalone insert shot to build tension, omitting any human characters from the frame.", "Applied skipped_close_framing rules by limiting the background description to the immediate frame-edge surfaces ([L07: interior wall...]) instead of referencing the entire room." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "a6d7e83dceb93412", "owned_hash": "80f679b464a34fb0", "camera_direction_hash": "70cf50605a58524f", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 12, "dependency_reason": "previous_shot_refs indicates S12_Shot20 as exact_background, though the background reference is mostly omitted here due to close framing on the door.", "_shot_index": 4, "visible_entities": [ "P05", "L04", "L07" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 27, "shot_index": 4 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "End the flow near the entrance with a close-up (CU), low angle, static through-gap composition on the old metal door about one meter away. The narrow opening sits just off-center, showing a sliver of exterior gray light while the door edge shudders in the wind, making the sound source feel like a presence pressing into the room.", "lighting_mood": "A thin blade of cold dusk light leaks through the metal door gap into otherwise gray-green interior gloom, turning the rattle into a visual threat.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "P05", "L04", "L07" ], "allowed_outlook_pairs": [ { "character_id": "C02", "outlook_id": "O03" }, { "character_id": "C04", "outlook_id": "O05" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "bg_living_kitchen_dusk_ransacked_windy", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "ransacked_rooftop_interior", "element_type": "environment_state", "character_name": "", "description": "The interior of an old rooftop room is left in a messy, ransacked state with scattered objects and debris lying around in broken layers. The entire space is heavily shadowed and enveloped in a cold, sickly green-gray gloom, dimly illuminated by the fading light of dusk.", "applies_to_shots": [ 1, 4 ], "source_facts": [ "Shot 1: The ransacked rooftop room stays visible behind her in broken layers", "Shot 1: Sickly interior green-gray mixes with dying dusk", "Shot 4: gray-green interior gloom", "Shot 4: cold dusk light leaks through" ], "visual_inferences": [ "The chaotic placement of objects from being ransacked remains untouched across the scene.", "The ambient light level is consistently dim, relying entirely on the minimal dusk light leaking from outside." ], "creative_decisions": [ "Maintain a consistent green-gray color grading to emphasize the bleak, empty, and anxious atmosphere of the scene, visually tying the isolated frames together." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 12, "shot_index": 20, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C02O03", "policy": "required" }, { "kind": "character_outlook", "id": "C04O05", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "5a0e8e35e186f887" }, { "scene_index": 28, "heading": "바다 + 갑판 / 해질 무렵", "beat_title": "사진 속 금월과 동요하는 시선", "representative_moment": "A young Asian woman's fingertip holds the worn edge of P08, revealing the text 'Geumwol' in the corner.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Extreme close-up, high-angle macro frame with shallow depth of field", "t2i_prompt": "Photorealistic cinematic still. Extreme close-up, high-angle macro composition hovering inches above [L19: wet, scratched fishing boat deck planks, darkened by sea spray]. Focus on P08, an old black-and-white photograph, held at the bottom edge by a young Asian woman's fingertip. The photograph faces the camera, featuring curled edges, worn paper fibers, salt stains, and a faint crimson-brown stain near the upper right corner where tiny handwritten text is barely visible. P12, a folded paper map, lies partially visible and out of focus at the frame edge. Cold dusk-blue ambient light pools on the damp wood and reflects on the photo's surface, with a shallow depth of field isolating the fingertip and the paper against the soft background.", "outfit_assignments": [], "source_facts": [ "S#28 action: 수리영 shows the photo", "shot description: an extreme close-up (ECU), high angle macro frame hovers inches above the wet deck, angled down onto the old black-and-white photograph", "shot description: tiny handwritten 'Geumwol' in the corner sits near the upper right, Suri-young's fingertip anchors the lower edge", "shot description: worn paper fibers, salt stains, curled photo corner emphasized with shallow DOF" ], "visual_inferences": [ "fixed_element dusk_ambient_lighting: Cold dusk-blue ambient light skims the damp deck", "P12 folded map is placed near the photo but out of focus to keep primary focus on P08", "A faint crimson-brown stain adds an occult charge to the photograph's texture" ], "creative_decisions": [ "Replaced C01O02 with 'young Asian woman's fingertip' as per isolated body-part focus rule", "Used L19 embedded description to anchor the background in close framing without referencing the chain_bg image", "High-angle macro perspective strictly confines the frame to the immediate interaction surface (deck, photo, map, and finger) to maintain single-camera scale" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "299c261f692a4b9e", "owned_hash": "36194228037de9ee", "camera_direction_hash": "f1e85ae5f15a8a38", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 25, "dependency_reason": "수리영이 꺼내어 보여주는 사진이 25씬에서 발견된 동일한 사진임", "_shot_index": 3, "visible_entities": [ "C01", "P08", "P12", "L19" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 28, "shot_index": 3 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Direct observer view derived from the low close-observation flow: an extreme close-up (ECU), high angle macro frame hovers inches above the wet deck, angled down onto the old black-and-white photograph. The tiny handwritten 'Geumwol' in the corner sits near the upper right of frame, while 수리영's fingertip anchors the lower edge at natural scale; the worn paper fibers, salt stains, and curled photo corner are emphasized with shallow DOF.", "lighting_mood": "Cold dusk-blue ambient light skims the damp deck, with a faint crimson-brown stain in the photograph corner giving the clue an occult charge.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "P08", "P12", "L19" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C03", "outlook_id": "O04" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "cb_deck_exterior_dusk_normal", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "dusk_ambient_lighting", "element_type": "environment_state", "character_name": "", "description": "The exterior of the small boat is enveloped in a cold, dusk-blue ambient light. This fading evening illumination creates a dim, moody atmosphere across the deck and exterior surfaces. The low light casts a desaturated, chilly tone over the surroundings, indicating the late hour on the water.", "applies_to_shots": [ 3, 8 ], "source_facts": [ "scene heading: 해질 무렵 (around dusk)", "Shot 3 lighting: Cold dusk-blue ambient light skims the damp deck", "Shot 8 lighting: Dying coastal dusk" ], "visual_inferences": [ "The overall ambient lighting remains consistently dim and bluish-gray across the boat's exterior during the short time frame of the scene." ], "creative_decisions": [ "Focusing on the cold dusk lighting as an environmental anchor to maintain visual consistency across drastically different shot scales." ], "confidence": "high" } ], "previous_shot_refs": [], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C03O04", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "c591e093acca7d72" }, { "scene_index": 28, "heading": "S#28. 바다 + 갑판 / 해질 무렵", "beat_title": "조타실을 몸으로 막아서는 수리영", "representative_moment": "수리영(C01)이 조타실 입구에서 두 팔을 벌려 인우(C03)를 가로막고 서 있는 순간.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Eye-level observer view from the deck, framing the doorway symmetrically to emphasize the foreground silhouette against the sickly interior light.", "t2i_prompt": "Photorealistic cinematic still. Focus on the tense standoff at the open wheelhouse entrance. C01O02 in a navy windbreaker, a young Asian woman, stands with her back to the camera, silhouetted against the sickly green-gray light spilling from the interior. Her full body is visible from boots to shoulders as she braces her arms wide across the sliding door from the reference, creating a human barricade. Beyond her, sharing the same doorway threshold, C03O04 in a green waterproof jacket, a young Asian man, is partially trapped inside the dim wheelhouse from the reference. He stands at a three-quarter left angle, halted at the helm console from the reference, his eyes fixed on C01O02. A cold, dusk-blue ambient light envelops the boat from the reference, casting a desaturated, chilly tone over the wet deck from the reference in the foreground.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C03", "outlook_id": "O04" } ], "source_facts": [ "S#28 action: '수리영, 조타실로 따라들어가 막으면서'", "shot description: 좁은 조타실 입구에서 두 팔을 벌린 채 인우의 앞을 막아선 수리영의 전신 구도.", "DP camera note: Direct observer view from just outside the wheelhouse entrance... frame holds 수리영 full-body in the center with both arms spread across the narrow opening." ], "visual_inferences": [ "fixed_element dusk_ambient_lighting: cold dusk-blue ambient light skims the damp deck", "수리영 is silhouetted against the interior spill light", "인우 is halted near the steering controls inside the wheelhouse" ], "creative_decisions": [ "Anchored background objects (boat, deck, sliding door, wheelhouse, helm console) directly to the reference image to prevent duplication", "Used the doorway as a shared foreground-background anchor to clearly establish spatial depth between the two characters", "Applied a color contrast between the sickly green-gray wheelhouse interior and the cold dusk-blue exterior to highlight the standoff" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "86a9394506c6c45f", "owned_hash": "36194228037de9ee", "camera_direction_hash": "d04156ebb659e32b", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "sliding door", "violating_phrase": "the sliding door from the reference", "reason": "This refers to an already existing door as an anchor, not to creating a new sliding door." }, { "owned_object": "helm console", "violating_phrase": "the helm console from the reference", "reason": "This refers to an already existing helm console as an anchor, not to creating a new helm console." }, { "owned_object": "boat", "violating_phrase": "the boat from the reference", "reason": "This refers to an already existing boat as an anchor, not to creating a new boat." }, { "owned_object": "deck", "violating_phrase": "the wet deck from the reference", "reason": "This refers to an already existing deck as an anchor, not to creating a new deck." }, { "owned_object": "wheelhouse", "violating_phrase": "the wheelhouse from the reference", "reason": "This refers to an already existing wheelhouse as an anchor, not to creating a new wheelhouse." } ] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "", "_shot_index": 8, "visible_entities": [ "C01", "C03", "P08", "P12", "L19" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 28, "shot_index": 8 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Direct observer view from just outside the wheelhouse entrance: a wide shot (WS), eye-level, static doorway frame holds 수리영 full-body in the center with both arms spread across the narrow opening. The camera stays far enough back to keep her boots, shoulders, and outstretched hands visible, while 인우 remains partially trapped beyond her in the dim wheelhouse, seen through the gap between her arm and the doorframe.", "lighting_mood": "Dying coastal dusk silhouettes 수리영 against sickly green-gray wheelhouse spill, turning her blocking posture into a tense human barricade.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "C03", "P08", "P12", "L19" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C03", "outlook_id": "O04" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "cb_deck_exterior_dusk_normal", "reference_usage": "exact_background", "owned_objects": [ "boat", "cleat", "deck", "deck hatch", "fishing rod rack", "fog", "helm console", "net bin", "rail", "rope coil", "sea", "sliding door", "wheelhouse", "winch", "window" ], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "owned_objects lists all environment objects already drawn in the background reference — do not create new ones", "when focusing on an owned object, anchor it explicitly: 'the door from the reference'" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "dusk_ambient_lighting", "element_type": "environment_state", "character_name": "", "description": "The exterior of the small boat is enveloped in a cold, dusk-blue ambient light. This fading evening illumination creates a dim, moody atmosphere across the deck and exterior surfaces. The low light casts a desaturated, chilly tone over the surroundings, indicating the late hour on the water.", "applies_to_shots": [ 3, 8 ], "source_facts": [ "scene heading: 해질 무렵 (around dusk)", "Shot 3 lighting: Cold dusk-blue ambient light skims the damp deck", "Shot 8 lighting: Dying coastal dusk" ], "visual_inferences": [ "The overall ambient lighting remains consistently dim and bluish-gray across the boat's exterior during the short time frame of the scene." ], "creative_decisions": [ "Focusing on the cold dusk lighting as an environmental anchor to maintain visual consistency across drastically different shot scales." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 28, "shot_index": 3, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C03O04", "policy": "required" }, { "kind": "background", "id": "cb_deck_exterior_dusk_normal", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "f753e800d0ac299a", "status": "contract_violation" }, { "scene_index": 29, "heading": "S#29. 조타실 / 해질 무렵", "beat_title": "절박한 진실의 토로", "representative_moment": "수리영(C01)이 눈가를 붉힌 채 엄마가 잡혀있다는 사실을 토로하며 절박한 표정으로 매달린다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Tight close-up, low angle, head-and-shoulder composition, dual-tone lighting", "t2i_prompt": "Photorealistic cinematic still. Low angle looking slightly up at C01O02 in a navy windbreaker, a young East Asian woman, leaning forward in a desperate plea at a three-quarter angle from the right. Her shoulders occupy the lower frame edge, her face positioned slightly left of center. Focus on her strained jaw muscles, slightly open mouth, and reddened eyes fixed intently on an unseen figure off-camera to the right. Sickly greenish-yellow fluorescent light falls across her forehead and cheek, mixing with a cold blue-gray dusk spill that washes the opposite side of her face and neck. [L20: A dark, out-of-focus wall surface softly compressing the frame immediately behind her shoulder, scattered with blurred dark crimson circular marks melting into a murky blue-gray haze].", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O02" } ], "source_facts": [ "Shot 3: tight close-up (CU), low angle, pushes in from slightly below her face", "Shot 3: reddened eyes, open mouth, and strained jaw muscles occupy most of the frame", "Shot 3: Sickly fluorescent interior light mixes with cold dusk spill" ], "visual_inferences": [ "fixed_element wheelhouse_dusk_lighting: dual-tone light consisting of greenish-yellow fluorescent and bruised blue-gray dusk spill", "fixed_element red_circular_wall_markings: represented as soft blurred dark crimson marks in the immediate background depth due to close framing", "She leans forward physically to match the tension of her pleading dialogue" ], "creative_decisions": [ "Used a head-and-shoulder composition rather than a screen-filling face to avoid floating head artifacts while maintaining the tight CU feel", "Applied shallow depth of field (skipped close framing) to isolate her expression, keeping background elements like the red circles visible only as soft color elements" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "db2f29ef4f309503", "owned_hash": "3dcdffa44980a305", "camera_direction_hash": "e478caefc8d825a4", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "No direct visual dependency on prior scenes for this specific isolated close-up.", "_shot_index": 3, "visible_entities": [ "C01", "P08", "L20" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 29, "shot_index": 3 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Direct observer view at 수리영's eyeline: a tight close-up (CU), low angle, pushes in from slightly below her face so her reddened eyes, open mouth, and strained jaw muscles occupy most of the frame without flattening into a frontal portrait. She is framed slightly left of center in three-quarter view, with the cramped wheelhouse melting into blue-gray blur behind her as she pleads about her mother.", "lighting_mood": "Sickly fluorescent interior light mixes with cold dusk spill, reddening her eyes while leaving the background in a bruised blue-gray haze.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "P08", "L20" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C03", "outlook_id": "O04" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "cb_wheelhouse_interior_dusk_red_circle_marks", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "wheelhouse_dusk_lighting", "element_type": "environment_state", "character_name": "", "description": "The cramped interior of the wheelhouse is illuminated by a constant, unsettling dual-tone light. A sickly, harsh greenish-yellow fluorescent light from the cabin fixtures mixes heavily with the cold, bruised blue-gray spill of dusk coming from outside, creating a murky and oppressive atmosphere.", "applies_to_shots": [ 3, 5, 7 ], "source_facts": [ "Shot 3: Sickly fluorescent interior light mixes with cold dusk spill, reddening her eyes while leaving the background in a bruised blue-gray haze.", "Shot 5: harsh greenish cabin light with cold blue dusk leaking", "Shot 7: Muted blue-gray cabin darkness... sickly fluorescent spill" ], "visual_inferences": [ "The ambient light sources remain constant throughout the short dialogue exchange, casting consistent color temperatures across different angles." ], "creative_decisions": [ "Defined the specific color palette (greenish-yellow vs blue-gray) as an environment state to ensure the claustrophobic and tense mood is maintained across all tight and macro shots." ], "confidence": "high" }, { "element_id": "red_circular_wall_markings", "element_type": "environment_state", "character_name": "", "description": "The dark, muted interior walls of the wheelhouse are densely covered with numerous crude, red circular markings arranged in rows. The crimson paint has a slightly wet or glossy texture that contrasts sharply against the grim, dimly lit background of the cabin.", "applies_to_shots": [ 3, 7 ], "source_facts": [ "Scene text: 조타실안 수많은 원... 붉은 원!", "Shot 7: dark wheelhouse wall fills the center and right with rows of red circular markings.", "Shot 7: sharp crimson of the circles... wall texture feel wet and real." ], "visual_inferences": [ "Although explicitly revealed as a focal point in Shot 7, these markings are physically painted on the walls and would exist in the background environment of Shot 3, even if obscured or heavily blurred by the shallow depth of field." ], "creative_decisions": [ "Excluded from Shot 5 as it is an extreme macro shot pointing downward at a photograph where walls would not be visible. Included in Shot 3 and 7 to ground the spatial continuity of the cabin." ], "confidence": "medium" } ], "previous_shot_refs": [], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C03O04", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "9da983ebfbcfeaab" }, { "scene_index": 29, "heading": "S#29. 조타실 / 해질 무렵", "beat_title": "사진 속 소녀가 엄마임을 밝히는 수리영", "representative_moment": "수리영의 흙 묻은 손가락이 낡은 흑백 사진 속 어린 소녀의 얼굴을 강하게 누르고 있다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "extreme close-up (ECU), high angle macro shot looking down, harsh greenish cabin light mixing with cold blue dusk", "t2i_prompt": "Photorealistic cinematic still. Extreme close-up, high angle macro shot looking down. Focus on the dirt-marked index finger of a young Asian woman pressing firmly against the printed face of a young Asian girl in the same old creased black-and-white photograph. The finger pad covers part of the printed features, mud stains contrasting against the faded photo paper. The photo border and [L20: a sliver of a shadowed, scratched counter surface] are visible at the frame edges for scale. The macro frame is illuminated by harsh greenish-yellow ambient light mixing with cold bruised blue-gray dusk spill, catching the grain of the photograph and making the dirt on the fingertip appear almost black.", "outfit_assignments": [], "source_facts": [ "S#29 action: '이 사진속의 얘가 우리 엄마예요.'", "shot description: 낡은 사진 속 어린 소녀의 얼굴을 꾹 누른 수리영의 흙 묻은 검지손가락 클로즈업.", "DP instruction: extreme close-up (ECU), high angle macro shot looks down at the photograph... compressing the frame to dirt-marked index finger pressing the young girl's face." ], "visual_inferences": [ "fixed_element wheelhouse_dusk_lighting: sickly greenish-yellow fluorescent light mixes heavily with cold, bruised blue-gray spill of dusk", "Applying id_policy.body_part_focus_rule: using common noun + demographic descriptor ('the dirt-marked index finger of a young Asian woman') instead of C01O02.", "Applying reproduction_surface_rule: using common noun ('printed face of a young Asian girl') instead of an entity ID for the face in the photo." ], "creative_decisions": [ "Used 'the same old creased black-and-white photograph' to maintain zoom_in_detail continuity without violating the skipped_close_framing background rule.", "Framed the shot to exclude surrounding hands and the room, using only the counter surface edge [L20] for scale, emphasizing the precise focal detail of the dirt-stained finger on the printed face." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "1f5845acffc69124", "owned_hash": "3dcdffa44980a305", "camera_direction_hash": "c68e2742a0a4a8e4", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 29, "dependency_reason": "zoom_in_detail on the photograph held in S29_Shot3, maintaining the exact dual-tone lighting (greenish-yellow and blue-gray dusk) while focusing on the macro detail of the finger pressing the photo.", "_shot_index": 5, "visible_entities": [ "C01", "P08", "L20" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 29, "shot_index": 5 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Direct observer view continuing the downward evidence move: an extreme close-up (ECU), high angle macro shot looks down at the photograph held between the two people, compressing the frame to 수리영's dirt-marked index finger pressing the young girl's face. The girl's face sits just right of center under the fingertip, while the surrounding hands are mostly excluded and only the photo border and a sliver of shadowed wheelhouse surface remain for scale.", "lighting_mood": "The macro frame is lit by harsh greenish cabin light with cold blue dusk leaking across the photo grain, making the fingertip's dirt look almost black.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C01", "P08", "L20" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C03", "outlook_id": "O04" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "cb_wheelhouse_interior_dusk_red_circle_marks", "reference_usage": "skipped_close_framing", "owned_objects": [], "camera_reference": null, "close_framing_skips_background_ref": true, "constraints": [ "this shot uses close framing — the background reference is intentionally omitted at composition time", "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list", "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent", "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from", "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "wheelhouse_dusk_lighting", "element_type": "environment_state", "character_name": "", "description": "The cramped interior of the wheelhouse is illuminated by a constant, unsettling dual-tone light. A sickly, harsh greenish-yellow fluorescent light from the cabin fixtures mixes heavily with the cold, bruised blue-gray spill of dusk coming from outside, creating a murky and oppressive atmosphere.", "applies_to_shots": [ 3, 5, 7 ], "source_facts": [ "Shot 3: Sickly fluorescent interior light mixes with cold dusk spill, reddening her eyes while leaving the background in a bruised blue-gray haze.", "Shot 5: harsh greenish cabin light with cold blue dusk leaking", "Shot 7: Muted blue-gray cabin darkness... sickly fluorescent spill" ], "visual_inferences": [ "The ambient light sources remain constant throughout the short dialogue exchange, casting consistent color temperatures across different angles." ], "creative_decisions": [ "Defined the specific color palette (greenish-yellow vs blue-gray) as an environment state to ensure the claustrophobic and tense mood is maintained across all tight and macro shots." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 29, "shot_index": 3, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C03O04", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "readiness_policy": "skipped_by_policy", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": true, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "55e471af9d34f555" }, { "scene_index": 29, "heading": "S#29. 조타실 / 해질 무렵", "beat_title": "붉은 원의 정체와 공포", "representative_moment": "인우의 어깨 너머로 조타실 벽면을 가득 채운 붉은 원형 표식들이 드러나는 순간.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Over-the-shoulder medium shot, eye-level, muted blue-gray dusk mixed with sickly fluorescent spill", "t2i_prompt": "Photorealistic cinematic still. Over-the-shoulder perspective. C03O04 in a green waterproof jacket, an East Asian man, stands with back and side toward the camera, his rigid hunched shoulder and a narrow slice of his cheek occupying the left foreground edge, eyes fixed on the wall. Focus on the dark wall panel from the reference filling the center and right frame, densely covered with rows of crude, glossy red circular markings. Muted blue-gray ambient dusk mixes with a sickly greenish-yellow ambient spill, catching the wet texture of the crimson paint. A dim blue dusk glow shines through the closed, salt-streaked window from the reference in the background.", "outfit_assignments": [ { "character_id": "C03", "outlook_id": "O04" } ], "source_facts": [ "S29 action: '사색이 된 인우. 인우 얼굴 너머에 보이는 조타실안 수많은 원.'", "Shot 7 description: 인우의 어깨 너머 조타실 어두운 벽면에 빼곡하게 칠해진 붉은색 원형 표식들의 정지 구도.", "DP instruction: wheelhouse side window is closed, salt-streaked, with dim dusk glow." ], "visual_inferences": [ "fixed_element wheelhouse_dusk_lighting: sickly harsh greenish-yellow fluorescent light mixing with cold blue-gray dusk spill.", "fixed_element red_circular_wall_markings: crimson paint has a slightly wet or glossy texture contrasting against the grim background." ], "creative_decisions": [ "Placed C03's shoulder strictly on the left edge to allow the ominous red circles to dominate the visual weight of the frame.", "Emphasized the wet, glossy texture of the paint under the sickly lighting to amplify the horror mood without showing blood." ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "ec76c833d6f380b4", "owned_hash": "3dcdffa44980a305", "camera_direction_hash": "05a44b010237961c", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "wall panel", "violating_phrase": "the wall panel from the reference filling the center and right frame", "reason": "This refers to an owned background object in a way that reinforces it as a newly depicted focal object with added detail, which goes beyond safe anchor-only reference." }, { "owned_object": "window", "violating_phrase": "a dim blue dusk glow shines through the closed, salt-streaked window from the reference", "reason": "This mentions the owned window with descriptive rendering language, implying the prompt is drawing that existing object again rather than only referencing it." } ] } } ], "scene_type": "normal", "dependent_scene_index": 29, "dependency_reason": "Shot 5에서 확립된 조타실 내부 조명과 분위기를 이어받아, 동일한 공간의 벽면(배경 참조)을 보여줌.", "_shot_index": 7, "visible_entities": [ "C03", "P08", "L20" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 29, "shot_index": 7 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Over-the-shoulder perspective behind 인우: a medium shot (MS), eye-level, static frame places his shoulder and a narrow slice of his cheek along the left foreground edge while the dark wheelhouse wall fills the center and right with rows of red circular markings. The composition holds completely still after the reveal, minimizing 인우 as a scale reference so the crimson circles become the occult focal point without cutting away from the established axis.", "lighting_mood": "Muted blue-gray cabin darkness is punctured by the sharp crimson of the circles, with just enough sickly fluorescent spill to make the wall texture feel wet and real.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "C03", "P08", "L20" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O02" }, { "character_id": "C03", "outlook_id": "O04" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "cb_wheelhouse_interior_dusk_red_circle_marks", "reference_usage": "exact_background", "owned_objects": [ "bench", "chair", "console", "dashboard", "display", "door", "fluorescent light", "radio microphone", "steering wheel", "storage locker", "switch panel", "throttle lever", "wall panel", "window", "windshield" ], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "owned_objects lists all environment objects already drawn in the background reference — do not create new ones", "when focusing on an owned object, anchor it explicitly: 'the door from the reference'" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "wheelhouse_dusk_lighting", "element_type": "environment_state", "character_name": "", "description": "The cramped interior of the wheelhouse is illuminated by a constant, unsettling dual-tone light. A sickly, harsh greenish-yellow fluorescent light from the cabin fixtures mixes heavily with the cold, bruised blue-gray spill of dusk coming from outside, creating a murky and oppressive atmosphere.", "applies_to_shots": [ 3, 5, 7 ], "source_facts": [ "Shot 3: Sickly fluorescent interior light mixes with cold dusk spill, reddening her eyes while leaving the background in a bruised blue-gray haze.", "Shot 5: harsh greenish cabin light with cold blue dusk leaking", "Shot 7: Muted blue-gray cabin darkness... sickly fluorescent spill" ], "visual_inferences": [ "The ambient light sources remain constant throughout the short dialogue exchange, casting consistent color temperatures across different angles." ], "creative_decisions": [ "Defined the specific color palette (greenish-yellow vs blue-gray) as an environment state to ensure the claustrophobic and tense mood is maintained across all tight and macro shots." ], "confidence": "high" }, { "element_id": "red_circular_wall_markings", "element_type": "environment_state", "character_name": "", "description": "The dark, muted interior walls of the wheelhouse are densely covered with numerous crude, red circular markings arranged in rows. The crimson paint has a slightly wet or glossy texture that contrasts sharply against the grim, dimly lit background of the cabin.", "applies_to_shots": [ 3, 7 ], "source_facts": [ "Scene text: 조타실안 수많은 원... 붉은 원!", "Shot 7: dark wheelhouse wall fills the center and right with rows of red circular markings.", "Shot 7: sharp crimson of the circles... wall texture feel wet and real." ], "visual_inferences": [ "Although explicitly revealed as a focal point in Shot 7, these markings are physically painted on the walls and would exist in the background environment of Shot 3, even if obscured or heavily blurred by the shallow depth of field." ], "creative_decisions": [ "Excluded from Shot 5 as it is an extreme macro shot pointing downward at a photograph where walls would not be visible. Included in Shot 3 and 7 to ground the spatial continuity of the cabin." ], "confidence": "medium" } ], "previous_shot_refs": [ { "scene_index": 29, "shot_index": 5, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C01O02", "policy": "required" }, { "kind": "character_outlook", "id": "C03O04", "policy": "required" }, { "kind": "background", "id": "cb_wheelhouse_interior_dusk_red_circle_marks", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "8b16f34e1bf512de", "status": "contract_violation" }, { "scene_index": 30, "heading": "S#30. 바다 / 해질 무렵", "beat_title": "안개 속으로 진입하는 어선", "representative_moment": "해 질 녘 황금빛 바다 위, 짙은 잿빛 안개벽을 향해 나아가는 작은 어선의 원경.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Extreme wide establishing shot from far offshore, cold gray-blue haze with fading gold dusk", "t2i_prompt": "Photorealistic cinematic still. Extreme wide establishing shot from far offshore, eye-level to the horizon. A small fishing boat is centered in the lower middle of the frame, its stern facing the camera, leaving a trail of wake. Focus on the towering gray fog from the reference that spans the entire horizon line ahead. The dark blue-gray sea from the reference stretches outward, catching a muted golden sunset reflection. Fading gold dusk is rapidly drained into a cold gray-blue haze, making the dense fog wall feel like a physical boundary dwarfing the boat.", "outfit_assignments": [], "source_facts": [ "scene text: 서서히 지는 해가 바다위에 황금빛 그림자를 드리운다. 세상과 경계를 지우는 것 같은 짙은 안개 띠. 서서히 안개 안으로 들어가는 인우의 배.", "DP instruction: Direct observer view from far offshore: an extreme wide / establishing shot (EWS), eye-level to the horizon", "DP instruction: The boat remains small and centered in the lower middle of frame, its bow aimed into a towering gray fog wall" ], "visual_inferences": [ "boat leaving a wake to indicate forward movement into the fog", "fog wall acting as a massive opaque barrier against the fading sunset" ], "creative_decisions": [ "Focus on the massive fog wall to emphasize the small scale and isolation of the boat", "Contrast the cold gray-blue haze with the muted golden reflection on the sea" ], "confidence": "high", "owned_validation": { "schema_version": 1, "t2i_prompt_hash": "95577df9021f24f3", "owned_hash": "8b48a58b29003f6d", "camera_direction_hash": "970ffee42e28799f", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "fog", "violating_phrase": "towering gray fog from the reference that spans the entire horizon line ahead", "reason": "Although it says 'from the reference', the prompt is still instructing the fog to be shown as a major scene element; however this is a reference-based mention of an already existing object, so it is not a clear new-draw instruction and does not violate the rules." }, { "owned_object": "sea", "violating_phrase": "The dark blue-gray sea from the reference stretches outward", "reason": "Although it mentions the sea, it explicitly anchors it as 'from the reference', which indicates the existing background sea rather than a new object to add." } ] } } ], "scene_type": "normal", "dependent_scene_index": 28, "dependency_reason": "uses exact_background reference (cb_deck_exterior_dusk_fog_wall) from the prior location to maintain the same sea and fog environment", "_shot_index": 2, "visible_entities": [ "L19" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 30, "shot_index": 2 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Direct observer view from far offshore: an extreme wide / establishing shot (EWS), eye-level to the horizon from about three meters above the water, tracks behind the small fishing boat on its stern-to-bow axis. The boat remains small and centered in the lower middle of frame, its bow aimed into a towering gray fog wall that spans the entire horizon line, while the sea surface and sunset reflection stretch outward to dwarf it.", "lighting_mood": "Fading gold dusk is rapidly drained into cold gray-blue haze, making the fog wall feel like a physical boundary between worlds.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces", "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges", "core_principles": [ "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden", "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden", "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together", "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden" ], "forbidden_combinations": [ "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject", "overhead high angle + feet visible at the lower frame edge with floor visible below them", "hip-height side camera + full silhouette including head and feet visible in same frame", "low angle looking up + ceiling and floor both fully visible in same frame" ], "recommended_phrasings": [ "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge", "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them", "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges", "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward" ], "self_check_steps": [ "identify the camera position phrasing (height + angle)", "list elements described inside the frame (foreground prop, body parts, background surfaces)", "verify each element's vertical/depth placement is geometrically consistent with that camera position", "if mismatch found, adjust frame-edge description to match camera position" ], "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과" }, "fg_bg_shared_anchor_rule": { "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)", "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space", "forbidden_phrasings": [ "fg/bg separation without naming the shared surface/space", "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)", "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share" ], "recommended_keywords": [ "shared bench / shared table / shared surface", "the same X they both occupy (X = bench, sofa, table, doorway, room)", "across the X from each other", " crosses the air between them (interaction prop visually links the two)", "the sharing a single line of sight from the camera" ], "self_check_steps": [ "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?", "is one foreground and the other background-separated?", "is the shared surface/space anchor explicitly named?", "if not, add one of the recommended keywords" ], "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제" }, "primary_framing_rule": { "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).", "framing_scale_keywords": { "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.", "close": [ "close-up", "CU", "MCU", "ECU", "XCU", "extreme close-up", "클로즈업", "손가락이", "손이", "눈이", "얼굴이" ], "wide": [ "wide shot", "establishing", "aerial", "전경", "전신" ], "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords" }, "close_framing_rules": { "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame", "other_entities_appearance_options": [ "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge", "soft background — out-of-focus silhouette via shallow depth of field", "absent — not rendered at all even if listed in visible_entities" ], "forbidden_combinations": [ "same character's full body and body-part close-up in the same frame (view mixing)", "two characters both rendered as sharp close-up faces in the same frame" ], "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "wide_medium_rules": { "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable", "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot", "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule", "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)" }, "self_check_steps": [ "identify framing_scale from camera_direction", "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent", "verify same character's full body + body-part close-up are not combined", "if close: verify two faces are not both rendered sharp" ], "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성" }, "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면." }, "constraints": [ "one prompt captures one still moment", "do not combine full-body framing and body-part close-up in one frame", "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule", "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule", "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule" ] }, "id_policy": { "allowed_base_entity_ids": [ "L19" ], "allowed_outlook_pairs": [], "must_use_composite_character_ids": true, "common_noun_required_when": [ "body-part close-up", "photo/poster/screen/mirror/reflection/projection depiction", "unregistered extra" ], "demographic_fallback_required": true, "face_identifiability_rule": { "use_entity_id_when": [ "face identifiable: front, profile, three-quarter, or eyes closed", "OTS framing with any visible face" ], "common_noun_required_when": [ "back to camera with no visible face", "silhouette or blurred outline only", "OTS with only back of head/shoulder, face entirely hidden" ], "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part", "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌" }, "body_part_focus_rule": { "trigger_phrases": [ "focus on", "close on", "tight on", "detail on" ], "applies_to": "any body part (face or non-face) when the framing is dominated by that part", "id_use": "forbidden — use a common noun + demographic descriptor instead", "common_noun_examples_generic": [ "a trembling hand", "an exposed forearm", "the furrowed brow of a figure in shadow", "the clenched jaw of a figure in profile" ] }, "close_framing_face_phrasing": { "forbidden_phrases": [ "face filling the entire frame", "face filling nearly the entire frame", "her face fills the frame", "shoulders only entering the frame", "head snapped sideways", "his face fills the frame" ], "recommended_phrasings": [ "tight framing on her face, upper body in frame", "close on his expression, shoulder visible at the edge", "head-and-shoulder composition, face occupying the center-right", "face in profile at the left, neck and upper chest visible" ], "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지" }, "reproduction_surface_rule": { "applies_to_surfaces": [ "photograph", "poster", "painting", "portrait", "monitor", "TV", "mirror", "window reflection", "projection" ], "id_use": "forbidden — common noun + demographic descriptor only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨" }, "demographic_descriptor_policy": { "required_on_first_appearance": true, "applies_to_id_forms": [ "C##", "C##O##" ], "token_count_range": [ 1, 2 ], "components": { "ethnicity": [ "Asian", "East Asian", "South Asian", "Southeast Asian", "Black", "Middle Eastern", "Hispanic", "Latina", "Latino", "Caucasian" ], "gender": [ "man", "woman", "figure" ], "age_band": [ "in his/her 20s", "in his/her 30s", "in his/her 40s", "young", "middle-aged", "elderly" ], "role_hint_from_outfit": [ "in worker uniform", "in fisher workwear", "in detective coat", "in business suit" ] }, "format_template_a": "C##O## in <옷 1-3 단어>, ", "format_template_b": "C##O##, , <자세 표현>", "source_priority": [ "visual_world_rules.region", "visual_world_rules.era", "shot context outfit hint" ], "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only", "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정" }, "constraints": [ "visible real people with outfit binding use C##O## composite IDs", "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead", "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)", "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')", "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns", "use the format `C##O## in <옷 1-3 단어>, ` or `C##O##, , <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "cb_deck_exterior_dusk_fog_wall", "reference_usage": "exact_background", "owned_objects": [ "cleat", "deck", "fishing rod rack", "fog", "hatch", "helm console", "net bin", "rail", "rope coil", "sea", "sliding door", "wheelhouse", "winch", "window" ], "camera_reference": null, "close_framing_skips_background_ref": false, "constraints": [ "owned_objects lists all environment objects already drawn in the background reference — do not create new ones", "when focusing on an owned object, anchor it explicitly: 'the door from the reference'" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [ { "scene_index": 28, "shot_index": 3, "ref_usage": "", "kind": "location" } ], "forward_zoom_targets": [], "cross_shot_id_substitution_rule": { "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs", "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##", "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures", "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체", "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라", "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다" }, "ref_usage_constraints": { "zoom_in_detail": { "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only", "forbidden_additions": [ "new persons not in the prior shot", "new props not in the prior shot", "new background elements not in the prior shot", "new poses, gestures, or expressions not in the prior shot", "different moments of action" ], "required_phrasings": [ "describe only the focused sub-region in concrete terms", "explicitly state the same environment (lighting, background) is preserved", "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'" ], "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)", "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사" }, "exact_background": { "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot", "permitted_additions": [ "new persons, poses, expressions, actions for this shot's moment", "different camera angle within the same room" ], "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)", "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt", "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관" }, "atmosphere_reference": { "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot", "forbidden_imports": [ "furniture layout from the prior shot", "wall positions from the prior shot", "specific architectural features from the prior shot" ], "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')", "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성" } }, "view_consistency": { "framing_scope_options": [ "third_person", "close_up" ], "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.", "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)", "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt", "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch", "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)", "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)", "mixing_forbidden": true, "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리" }, "constraints": [ "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented", "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt", "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments", "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room", "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone", "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "background", "id": "cb_deck_exterior_dusk_fog_wall", "policy": "required" } ], "forbidden_refs": [], "readiness_policy": "block_if_missing", "constraints": [ "do not imply or describe a reference image that is not listed in required_refs (no phantom references)", "if a required character_outlook ref is missing, scene image generation must block before the image stage" ] }, "_card_metadata": { "lift_status": { "rule_a_lifted": true, "rule_c_lifted": true, "rule_e_lifted": false, "rule_h_lifted": true, "id_policy_composite_lifted": true, "id_policy_close_framing_face_lifted": true, "id_policy_reproduction_surface_lifted": true, "continuity_cross_shot_id_substitution_lifted": true, "continuity_ref_usage_constraints_lifted": true, "continuity_view_consistency_lifted": true, "continuity_constraints_extended": true, "spatial_camera_frame_lifted": true, "spatial_fg_bg_shared_anchor_lifted": true, "spatial_primary_framing_lifted": true, "spatial_constraints_extended": true }, "rule_source": { "camera_rule": "A", "owned_rule": "C", "close_skip_rule": "E", "demographic_rule": "H", "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)", "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)", "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)", "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section", "ref_usage_rule": "v20/scene_detail/system.md#continuity-section", "view_consistency_rule": "v20/scene_detail/system.md#continuity-section", "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency", "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency", "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency" } } }, "render_prompt_card_hash": "d183beff419a3d10", "status": "contract_violation" } ] }, "step_id": "scene_detail", "run_id": "199e4c34-47c6-4200-89c3-a43e51f4223a", "resolved_model": "gemini-pro", "updated_at": "2026-05-05T07:17:32.009521+00:00", "project_config_snapshot": {} }