{ "status": "completed", "completed_count": 54, "applicable_count": 54, "failed_count": 0, "schema_version": 13, "config_hash": "85e027dca53bb4b3", "data": { "scenes": [ { "scene_index": 1, "heading": "S#1. 숲속 / 해 질 무렵", "beat_title": "검은 인영의 습격", "representative_moment": "C08O08이 도망치던 희생자의 등 위로 덮쳐 누르는 역동적인 측면 충돌 순간.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Wide side-profile at waist height with cool desaturated dusk lighting", "t2i_prompt": "Photorealistic cinematic still. Wide side-profile composition captured from waist height. C08O08 in a dark long coat, a Korean adult man with black hair, airborne in the upper-left space at three-quarter angle from the right, looking down, exerting heavy downward force onto the back of a severely injured Korean adult man in hiking clothes. The victim, positioned in the right half of the frame, pitching forward mid-stride in right profile, looking down with a pained expression, his thigh visibly wounded with a deep-crimson blood accent. Both figures physically locked together mid-collision above a shared dirt path. Dense overgrown bushes establishing the escape route behind them in the background. Dusk ambient light filtering through the trees in a cool desaturated palette with restrained low-key contrast, focusing on the physical tension of the airborne downward strike.", "outfit_assignments": [ { "character_id": "C08", "outlook_id": "O08" } ], "source_facts": [ "S#1 action: '어디선가 쏜살같이 달려온 검은 인영이 남자를 덮친다'", "shot description: airborne figure pressing onto the victim's back, mid-stride collision" ], "visual_inferences": [ "victim pitching forward from the impact weight", "dark long coat and hiking clothes reacting to the dynamic motion" ], "creative_decisions": [ "waist height perspective to cleanly frame both torsos and the dramatic angle of impact in wide framing", "deep-crimson accent on the thigh wound contrasting sharply with the cool desaturated dusk ambient" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "5acaabb3af29b09e", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "e6b8433db45ab07e", "owned_usage_hash": "4f53cda18c2baa0c", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "No prior visual dependency.", "_shot_index": 3, "visible_entities": [ "L01", "C06", "C08" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 1, "shot_index": 3 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "wide", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At waist height and several meters to the side, the camera tracks parallel in a broad profile, widening only enough to preserve the changed distance and reveal the entire collision. 금월산 남성 피해자 pitches across the right half of frame while 검은 형상의 존재 descends through the upper-left space onto his back, leaving visible lateral room for the full forward stagger.", "lighting_mood": "Dusk ambient light is treated in a cool desaturated palette with restrained contrast and the victim's blood held as a deep-crimson accent.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": null, "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": [ "L01", "C06", "C08" ], "allowed_outlook_pairs": [ { "character_id": "C06", "outlook_id": "O06" }, { "character_id": "C08", "outlook_id": "O08" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [ { "subject_id": "C06", "policy_type": "identity_reference", "policy": "generic_descriptor_allowed", "reason": "visible_shot_count=1 — selected-shot 저빈도 주변 subject" } ], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "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" ] }, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C08O08", "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" ] }, "render_contracts": [], "_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": "634e35f81c9e5b99" }, { "scene_index": 2, "heading": "S#2. 도시 뒷골목 / N", "beat_title": "물어뜯긴 목덜미와 붉은 눈동자", "representative_moment": "골목 전봇대 아래, 붉은 눈을 번뜩이는 C09가 여자의 목덜미를 물어뜯는 익스트림 클로즈업.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "extreme close-up from an oblique angle, restrained low-key lighting with selective crimson accent", "t2i_prompt": "Photorealistic cinematic still. Extreme close-up from an oblique angle. C09O09 in a dark long coat, an adult Korean man, at three-quarter angle from the left, his glowing red eye occupying most of the central field, gazing off-screen. The figure exerting force has his jaw locked firmly into the neck of an adult Korean woman in an everyday dress, seen from an over-shoulder angle along the lower edge of the frame, severely injured with a pained expression. Sharp teeth are visible tearing into the flesh at the contact point. [: A narrow peripheral strip of a concrete utility pole] anchors the entwined bodies in the background. Faint streetlight shapes the faces in restrained low-key contrast within a cool desaturated palette, isolating the man's red iris as the strongest crimson accent.", "outfit_assignments": [ { "character_id": "C09", "outlook_id": "O09" } ], "source_facts": [ "S#2 action: 남자의 붉은 눈동자, 목의 살점을 뜯는 이빨", "shot description: extreme close-up of the stranger's red eye, teeth biting into the victim's neck", "shot staging camera direction: extreme close-up in which his red eye occupies most of the central field while his teeth and the woman's neck remain legible along the lower edge" ], "visual_inferences": [ "oblique angle keeps the image predatory rather than portrait-like", "faint streetlight provides the explicitly faint illumination falling across the close observation" ], "creative_decisions": [ "use of L02 block to represent the background utility pole fragment without invoking the forbidden L01 ID", "cool desaturated color temperature to amplify the isolation of the crimson eye" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "a99232c6f1c7e017", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "e0a97fc852ab65c4", "owned_usage_hash": "4f53cda18c2baa0c", "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": [ "C07", "C09" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 2, "shot_index": 2 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At the man's eye height and slightly beside his face, the continuing dolly-in resolves into an extreme close-up in which his red eye occupies most of the central field while his teeth and the woman's neck remain legible along the lower edge. The oblique angle keeps the image predatory rather than portrait-like, with the utility pole and alley reduced to narrow peripheral fragments.", "lighting_mood": "Faint streetlight shapes the faces in restrained low-key contrast, with the man's red iris isolated as the strongest crimson accent in the cool desaturated image.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": null, "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", "C09" ], "allowed_outlook_pairs": [ { "character_id": "C07", "outlook_id": "O07" }, { "character_id": "C09", "outlook_id": "O09" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [ { "subject_id": "C07", "policy_type": "identity_reference", "policy": "generic_descriptor_allowed", "reason": "visible_shot_count=1 — selected-shot 저빈도 주변 subject" } ], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "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", "describe the close subject and its immediate surroundings without relying on the absent background reference", "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C09O09", "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" ] }, "render_contracts": [], "_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": "1e548c7245a23009" }, { "scene_index": 3, "heading": "S#3. 현장 / D", "beat_title": "뉴스 화면 속 현장을 지휘하는 혜수", "representative_moment": "폴리스 라인이 쳐진 숲속 현장에서 C02가 화면의 우측에 치우쳐 선 채로 날카롭게 프레임 밖을 응시하고 있다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "chest-height medium shot simulating local news footage, cool desaturated palette", "t2i_prompt": "Photorealistic cinematic still. Captured from an observer camera at chest height, simulating local-news footage with a cool desaturated broadcast palette and raw documentary texture. C02O02 in a dark leather jacket, a Korean middle-aged woman with a mature oval face and black hair, occupies the right-center of the frame in three-quarter left profile. She is leaning forward with shoulders set, looking off-screen toward the left edge. A yellow police line runs diagonally behind her as a shallow boundary. In the soft out-of-focus background, unnamed figures in police jackets are processing the forest scene among the trees, subordinate to her focus. Daylight illumination with restrained contrast and slight digital noise.", "outfit_assignments": [ { "character_id": "C02", "outlook_id": "O02" } ], "source_facts": [ "S#3 action: '씬1, 2 현장을 감식하는 경찰, 폴리스 라인을 치고 시신을 수거한다.'", "shot description: 혜수의 상체가 폴리스 라인 근처에 선 채 날카로운 시선으로 한 곳을 응시함" ], "visual_inferences": [ "forest location inferred from S1_Shot3 reference (previous scene site)", "police processing the scene placed behind the diagonal police line to emphasize her command position", "broadcast texture and slight digital noise inferred from 'local-news footage'" ], "creative_decisions": [ "chest-height medium framing holding her upper body in the right-center to create spatial tension with her leftward gaze", "cool desaturated daylight palette and restrained contrast to match the raw documentary tone" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "0add3bb366d5a4ae", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "ff18ffa2bef1638f", "owned_usage_hash": "4f53cda18c2baa0c", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 1, "dependency_reason": "S1_Shot3의 숲속 사건 현장 배경을 참조하여 뉴스 화면 속 현장의 시각적 연속성을 유지하기 위함", "_shot_index": 2, "visible_entities": [ "L01", "C02" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 3, "shot_index": 2 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Presented as local-news footage, the camera settles at 혜수's chest height near the police line after its dolly-in, holding her upper body in the right-center at a slight three-quarter angle. Her fixed eyeline cuts toward an unseen area beyond the left edge while the police line remains as a shallow diagonal boundary behind her.", "lighting_mood": "Daylight appropriate to the location is rendered with a cool desaturated broadcast image, restrained contrast, and raw documentary texture.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": null, "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": [ "L01", "C02" ], "allowed_outlook_pairs": [ { "character_id": "C02", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "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" ] }, "continuity_elements_used": { "fixed_elements": [], "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C02O02", "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" ] }, "render_contracts": [], "_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": "ac500135bbe06656" }, { "scene_index": 4, "heading": "S#4. 옥탑방 / D", "beat_title": "고립된 옥탑방 전경", "representative_moment": "높은 부감으로 내려다본 허름한 옥탑방과 낡은 빌라 옥상의 쓸쓸한 전경.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Omniscient pronounced overhead angle, cool desaturated daylight", "t2i_prompt": "Photorealistic cinematic still. The shabby rooftop room is positioned in the middle-center midground. From a pronounced overhead high angle looking down, the centered rooftop room occupies less than the central third of the frame, surrounded by the broad base of the old villa roof. Cool, desaturated daylight casts restrained low-key shadows across the aged exterior, highlighting tactile architectural details. The layout of the rooftop, anchored by a weathered parapet, a water tank, and a utility pole with tangled wire, establishes a strong sense of isolation. The camera focuses on the bleak isolation of the rooftop room against the surrounding urban decay from above.", "outfit_assignments": [], "source_facts": [ "S#4 text: '인천 변두리 부감, 오래된 빌라 옥탑방 전경.'", "shot description: '낡은 다세대 빌라 옥상에 자리한 허름한 옥탑방을 높은 곳에서 내려다보는 부감 구도'" ], "visual_inferences": [ "Voiceover discussing unresolved crimes implies a somber, isolated, and suspenseful mood for the setting.", "The urban decay elements like the parapet, water tank, and utility pole naturally fit the L03 background reference." ], "creative_decisions": [ "Use a pronounced overhead crane angle to center the room and emphasize its isolation relative to the surrounding roof layout.", "Apply a cool, desaturated daylight grading with restrained contrast to match the overarching occult thriller tone and heighten the sense of unease." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001" ], "reference_phrase_kinds": [ "background" ], "owned_object_usage": [ { "owned_token": "AC", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "antenna", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "car", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "clothesline", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "gas meter", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "laundry", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "parapet", "usage_kind": "anchor", "source_phrase": "anchored by a weathered parapet" }, { "owned_token": "pipe", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "plant pot", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "rooftop", "usage_kind": "anchor", "source_phrase": "The layout of the rooftop" }, { "owned_token": "rooftop room", "usage_kind": "anchor", "source_phrase": "The shabby rooftop room is positioned in the middle-center midground." }, { "owned_token": "satellite dish", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "security grille", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "storage bin", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "utility pole", "usage_kind": "anchor", "source_phrase": "a utility pole with tangled wire" }, { "owned_token": "villa", "usage_kind": "anchor", "source_phrase": "surrounded by the broad base of the old villa roof" }, { "owned_token": "water tank", "usage_kind": "anchor", "source_phrase": "a water tank" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wire", "usage_kind": "anchor", "source_phrase": "tangled wire" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "c64935b5c16fa55c", "owned_hash": "363a68529d08338e", "camera_direction_hash": "cbcdd80cdbe3c991", "owned_usage_hash": "9a162bc4f67d9612", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "rooftop room", "violating_phrase": "The shabby rooftop room is positioned in the middle-center midground.", "reason": "The prompt mentions the rooftop room to describe its placement and composition within the frame as a spatial anchor, which matches the producer's 'anchor' declaration.", "verdict": "anchor_reference" }, { "owned_object": "villa", "violating_phrase": "surrounded by the broad base of the old villa roof", "reason": "The old villa roof is referenced solely to establish the spatial composition and surrounding environment, matching the producer's 'anchor' declaration.", "verdict": "anchor_reference" }, { "owned_object": "rooftop", "violating_phrase": "The layout of the rooftop", "reason": "The layout of the rooftop is described to set the scene structure and camera framing, aligning with the producer's 'anchor' declaration.", "verdict": "anchor_reference" }, { "owned_object": "parapet", "violating_phrase": "anchored by a weathered parapet", "reason": "The parapet is mentioned as a compositional landmark to establish scale and mood, which aligns with the producer's 'anchor' declaration.", "verdict": "anchor_reference" }, { "owned_object": "water tank", "violating_phrase": "a water tank", "reason": "The water tank is referenced as a scenic element to anchor the layout of the shot, matching the producer's 'anchor' declaration.", "verdict": "anchor_reference" }, { "owned_object": "utility pole", "violating_phrase": "a utility pole with tangled wire", "reason": "The utility pole is mentioned as part of the existing environment to anchor the composition, aligning with the producer's 'anchor' declaration.", "verdict": "anchor_reference" }, { "owned_object": "wire", "violating_phrase": "tangled wire", "reason": "The tangled wire is referred to as an environmental detail attached to the utility pole, which matches the producer's 'anchor' declaration.", "verdict": "anchor_reference" } ] } } ], "scene_type": "voiceover", "dependent_scene_index": -1, "dependency_reason": "앞쪽 시각 연관 씬 없음", "_shot_index": 1, "visible_entities": [ "L03" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 4, "shot_index": 1 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "wide", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From high above the building at a pronounced overhead angle, the crane descent reaches its end position with the rooftop room centered and the surrounding villa roof still clearly legible. The room occupies less than the central third, allowing the roof layout and the building's aged exterior to establish its isolation without distorting scale.", "lighting_mood": "Daylight appropriate to the setting is graded cool and desaturated with restrained contrast and tactile architectural detail.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "required_background_position", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "midground", "gesture_action": "none", "gesture_target_label": "", "label": "shabby rooftop room", "screen_zone": "middle_center", "target_id": "", "target_kind": "background" } ] }, "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": [ "L03" ], "allowed_outlook_pairs": [], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L03B02", "reference_usage": "exact_background", "owned_objects": [ "AC", "antenna", "car", "clothesline", "door", "gas meter", "laundry", "parapet", "pipe", "plant pot", "rooftop", "rooftop room", "satellite dish", "security grille", "storage bin", "utility pole", "villa", "water tank", "window", "wire" ], "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "background", "id": "L03B02", "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" ] }, "render_contracts": [], "_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": "3cfe2b7e5aa7fe60" }, { "scene_index": 5, "heading": "S#5. 옥탑방 안 / 실내", "beat_title": "엄마의 체취와 안도감", "representative_moment": "C01 leans in with her eyes closed to smell C03's shoulder by the entrance door, while C03 looks down with a distinct circular mark visible on her hanging wrist.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Observer POV at upper-torso height, close medium composition with a cool desaturated grade and selective gold accent.", "t2i_prompt": "Photorealistic cinematic still. C01O01 in a casual hooded jumper, a young Korean woman, stands in the middle-left foreground leaning toward her mother. She is in right profile with eyes closed, leaning her nose to the shoulder of C03O03 in comfortable homewear, a middle-aged Korean woman, positioned in the middle-right midground beside the entrance door. C03O03 is seen at a three-quarter angle from the left, looking down, her weight settled near the doorway. The shared entrance door space anchors the two figures in a close medium composition, captured at upper-torso height showing them from the waist up. A distinct, small circular mark is clearly imprinted on the bare skin of C03O03's wrist; her lower arm hangs relaxed and unobstructed beside her torso in the lower frame. Soft, restrained daylight filters through a small window in the background, wrapping the domestic interior space in a cool, desaturated color palette with low contrast and a faint selective gold accent.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C03", "outlook_id": "O03" } ], "source_facts": [ "딸, 수리영이 가벼운 차림새로 신발을 신으며 엄마의 냄새를 맡는다.", "수리영: profile_right, leaning nose to shoulder, gaze=closed_eyes", "강민숙: three_quarter_left, weight settled near doorway, gaze=down", "힐끗, 민숙 손목에 보이는 작은 원 표식.", "거실에 난 작은 창으로 햇살에 가늘게 비친다." ], "visual_inferences": [ "fixed_element minsuk_wrist_mark: wrist and lower arm hang relaxed and unobstructed beside her torso.", "fixed_element interior_daylight_lighting: cool desaturated color palette with low contrast." ], "creative_decisions": [ "Used a waist-up medium framing from an upper-torso height to capture both the intimate shoulder-leaning gesture and the hanging wrist mark in a single consistent geometric frame.", "Explicitly referenced the entrance door as a shared spatial anchor to ensure proper foreground/midground depth positioning between the two characters." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002" ], "reference_phrase_kinds": [ "character", "background" ], "owned_object_usage": [ { "owned_token": "TV", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "TV stand", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "cabinet", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "counter", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "dining table", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "anchor", "source_phrase": "beside the entrance door" }, { "owned_token": "sink", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "stovetop", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "anchor", "source_phrase": "through a small window in the background" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "54472362f04e9d52", "owned_hash": "6ccc14404189806b", "camera_direction_hash": "cd3756ccd8f03897", "owned_usage_hash": "83d213b7f907807a", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "door", "violating_phrase": "beside the entrance door", "reason": "The door is used solely as a spatial anchor to position the characters and contains no instructions to rebuild, repaint, or modify its appearance, matching the producer's anchor declaration.", "verdict": "anchor_reference" }, { "owned_object": "window", "violating_phrase": "filters through a small window in the background", "reason": "The window in the background serves purely as a light source to establish the scene's lighting, with no instructions to alter or redraw the window itself, matching the producer's anchor declaration.", "verdict": "anchor_reference" } ] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "", "_shot_index": 5, "visible_entities": [ "L05", "C01", "C03" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 5, "shot_index": 5 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At upper-torso height about one meter away, the camera completes its short arc toward 강민숙's shoulder side and holds the two women in a close medium composition. 수리영 leans in from the left with her nose near her mother's shoulder, while 강민숙 remains on the right near the entrance and her arm is left unobstructed for the impending downward move.", "lighting_mood": "Sunlight from the small living-room window is kept soft and restrained within a cool desaturated grade, with a faint selective gold accent preserving the domestic tenderness.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "shared_space_relation", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "foreground", "gesture_action": "none", "gesture_target_label": "", "label": "수리영 leaning toward her mother", "screen_zone": "middle_left", "target_id": "C01", "target_kind": "character" }, { "constraint_id": "fsc_002", "depth_plane": "midground", "gesture_action": "none", "gesture_target_label": "", "label": "강민숙 beside the entrance", "screen_zone": "middle_right", "target_id": "C03", "target_kind": "character" } ] }, "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": [ "L05", "C01", "C03" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C03", "outlook_id": "O03" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L05B02", "reference_usage": "exact_background", "owned_objects": [ "TV", "TV stand", "cabinet", "chair", "counter", "dining table", "door", "sink", "stovetop", "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "minsuk_wrist_mark", "element_type": "character_state", "character_name": "강민숙", "description": "A distinct, small circular mark is clearly imprinted on the visible side of the 40-year-old woman's wrist. The wrist and lower arm hang relaxed and unobstructed beside her torso, maintaining the exact position and visibility of the mark.", "applies_to_shots": [ 5, 6 ], "element_scope": "close", "source_facts": [ "힐끗, 민숙 손목에 보이는 작은 원(이하 표식)을 본다.", "오늘따라 유난히 또렷해 보이지만", "The circular mark sits near center and occupies less than a third of the frame", "her arm is left unobstructed for the impending downward move" ], "visual_inferences": [ "The mark is a clean geometric circle rather than an irregular scar, situated distinctly on the bare skin of the lower forearm or wrist." ], "creative_decisions": [ "The arm remains hanging loosely to keep the wrist prominently visible and stationary across both the medium shot and the extreme close-up." ], "confidence": "high" }, { "element_id": "interior_daylight_lighting", "element_type": "environment_state", "character_name": "", "description": "The domestic interior space is illuminated by soft, restrained daylight filtering through a small window. The overall environment maintains a cool, desaturated color palette with low contrast, preventing any harsh shadows.", "applies_to_shots": [ 5, 6 ], "element_scope": "full", "source_facts": [ "거실에 난 작은 창으로 햇살에 가늘게 비친다.", "Sunlight from the small living-room window is kept soft and restrained within a cool desaturated grade", "Established interior daylight renders the wrist in a cool desaturated palette" ], "visual_inferences": [ "The ambient light softly wraps around the subjects, keeping background details dim but recognizable." ], "creative_decisions": [ "Maintaining a cool, slightly desaturated grade to emphasize the distinctive wrist mark and preserve a calm, tender morning 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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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": "C03O03", "policy": "required" }, { "kind": "background", "id": "L05B02", "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" ] }, "render_contracts": [], "_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": "a49c6b4b794c94a6" }, { "scene_index": 5, "heading": "S#5. 옥탑방 안 / 실내", "beat_title": "민숙의 손목 표식을 보는 수리영", "representative_moment": "The camera holds static at wrist height, capturing a distinct circular mark on the wrist of C03.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "static insert framing at wrist height, cool desaturated grade with low contrast", "t2i_prompt": "Photorealistic cinematic still. The marked wrist of C03 occupies the middle-center foreground. A distinct, small circular mark is clearly imprinted on the visible side of the wrist. C03 in a casual sleeve, a Korean early-40s woman, stands at three-quarter angle from the left, her wrist and lower arm hanging relaxed and unobstructed beside her torso. The camera holds static at wrist height, aligned squarely with the mark, retaining a narrow portion of her forearm as an anatomical scale reference, while her head and lower legs are implied beyond the frame edges. The domestic interior space behind her is illuminated by soft, restrained daylight filtering through an out-of-focus window. The overall environment maintains a cool, desaturated color palette with low contrast, preventing any harsh shadows and keeping the background details dim.", "outfit_assignments": [ { "character_id": "C03", "outlook_id": "O03" } ], "source_facts": [ "힐끗, 민숙 손목에 보이는 작은 원(이하 표식)을 본다.", "오늘따라 유난히 또렷해 보이지만", "The circular mark sits near center and occupies less than a third of the frame", "her arm is left unobstructed for the impending downward move" ], "visual_inferences": [ "fixed_element minsuk_wrist_mark: wrist and lower arm hang relaxed and unobstructed beside her torso", "fixed_element interior_daylight_lighting: cool, desaturated color palette with low contrast", "insert framing excludes the face, focusing entirely on the physical mark and the lower arm geometry" ], "creative_decisions": [ "Positioned the camera directly at wrist height to execute the insert framing correctly without capturing the head or feet, complying with the spatial camera frame rules.", "Softened the background into a shallow depth of field, using the window purely as a lighting anchor to emphasize the stark clarity of the small mark on the skin." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001" ], "reference_phrase_kinds": [ "character", "background" ], "owned_object_usage": [ { "owned_token": "TV", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "TV stand", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "cabinet", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "counter", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "dining table", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "sink", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "stovetop", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "anchor", "source_phrase": "daylight filtering through an out-of-focus window" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "2fab2cd6792d1847", "owned_hash": "6ccc14404189806b", "camera_direction_hash": "a00c83cd74529dd4", "owned_usage_hash": "4269dfa96420888f", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "window", "violating_phrase": "daylight filtering through an out-of-focus window", "reason": "The prompt mentions the window as an out-of-focus anchor element for background lighting, which is not a redraw instruction, aligning with the producer's anchor declaration.", "verdict": "anchor_reference" } ] } } ], "scene_type": "normal", "dependent_scene_index": 5, "dependency_reason": "Preserves exact room background and lighting mood from Shot 5 while tightening the frame onto the wrist.", "_shot_index": 6, "visible_entities": [ "C03" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 5, "shot_index": 6 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "insert", "moment_lock": "single still moment from the selected shot only", "camera_direction": "The camera has lowered along 강민숙's arm and now holds static at wrist height, aligned squarely with the visible side of her wrist. The circular mark sits near center and occupies less than a third of the frame, with the wrist and a narrow portion of her forearm retained as anatomical scale references; the sole emphasized change from the previous shot is the drastic reduction in camera distance.", "lighting_mood": "Established interior daylight renders the wrist in a cool desaturated palette, with restrained contrast making the small circular mark unusually distinct.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "primary_subject_isolation", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "foreground", "gesture_action": "none", "gesture_target_label": "", "label": "marked wrist", "screen_zone": "middle_center", "target_id": "C03", "target_kind": "character" } ] }, "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" ], "allowed_outlook_pairs": [ { "character_id": "C03", "outlook_id": "O03" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [ { "subject_id": "C03", "policy_type": "identity_reference", "policy": "base_id_required", "reason": "Only 강민숙's wrist and a narrow portion of her forearm are visible, so an outlook-level reference would not align reliably with the insert." } ], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L05B02", "reference_usage": "exact_background", "owned_objects": [ "TV", "TV stand", "cabinet", "chair", "counter", "dining table", "door", "sink", "stovetop", "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "minsuk_wrist_mark", "element_type": "character_state", "character_name": "강민숙", "description": "A distinct, small circular mark is clearly imprinted on the visible side of the 40-year-old woman's wrist. The wrist and lower arm hang relaxed and unobstructed beside her torso, maintaining the exact position and visibility of the mark.", "applies_to_shots": [ 5, 6 ], "element_scope": "close", "source_facts": [ "힐끗, 민숙 손목에 보이는 작은 원(이하 표식)을 본다.", "오늘따라 유난히 또렷해 보이지만", "The circular mark sits near center and occupies less than a third of the frame", "her arm is left unobstructed for the impending downward move" ], "visual_inferences": [ "The mark is a clean geometric circle rather than an irregular scar, situated distinctly on the bare skin of the lower forearm or wrist." ], "creative_decisions": [ "The arm remains hanging loosely to keep the wrist prominently visible and stationary across both the medium shot and the extreme close-up." ], "confidence": "high" }, { "element_id": "interior_daylight_lighting", "element_type": "environment_state", "character_name": "", "description": "The domestic interior space is illuminated by soft, restrained daylight filtering through a small window. The overall environment maintains a cool, desaturated color palette with low contrast, preventing any harsh shadows.", "applies_to_shots": [ 5, 6 ], "element_scope": "full", "source_facts": [ "거실에 난 작은 창으로 햇살에 가늘게 비친다.", "Sunlight from the small living-room window is kept soft and restrained within a cool desaturated grade", "Established interior daylight renders the wrist in a cool desaturated palette" ], "visual_inferences": [ "The ambient light softly wraps around the subjects, keeping background details dim but recognizable." ], "creative_decisions": [ "Maintaining a cool, slightly desaturated grade to emphasize the distinctive wrist mark and preserve a calm, tender morning atmosphere." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 5, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character", "id": "C03", "policy": "required" }, { "kind": "background", "id": "L05B02", "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" ] }, "render_contracts": [], "_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": "61b5fac1b626b157" }, { "scene_index": 6, "heading": "S#6. 해안가 도로 / D", "beat_title": "활기찬 자전거 주행", "representative_moment": "해안가 도로에서 자전거 페달을 밟으며 달리는 C01의 측면 샷.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "At torso height beside the bicycle, tracking parallel at a stable medium lateral distance, cool desaturated daylight", "t2i_prompt": "Photorealistic cinematic still. C01O01 in a casual hooded jacket, an Asian young woman, occupies the middle-left midground, moving toward the road ahead. Seen at torso height from a stable medium lateral distance, in right profile, gazing into the distance. She is seated on the saddle of the bicycle in mid-pedal, one knee bent driving the pedal downward, the bicycle frame and wheels clearly visible beneath her. Her long black tied hair and jacket flap stream backward in the direction opposite to motion. She wears earphones. L06, the horizontal coastal road, runs behind and beneath her, with the adjacent coastal waters serving as a secondary lateral context. Lit by cool desaturated daylight with restrained contrast, revealing crisp tactile details in the moving hair and clothing.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "S#6 action: 수리영 riding a bicycle on the coastal road, wearing earphones, looking bright and energetic", "shot description: mid-pedal down stroke, trailing hair and clothing in right profile" ], "visual_inferences": [ "motion freeze captured by streaming hair and jacket flap", "torso height medium lateral tracking shot captures both the rider's energetic movement and the bicycle frame", "the road and coast provide the lateral background context" ], "creative_decisions": [ "cool desaturated daylight palette to balance the energetic motion with the project's overall tone", "middle-left framing to emphasize the forward movement direction toward the right" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001" ], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "584dce54b91f7912", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "58eccc1337d139c6", "owned_usage_hash": "4f53cda18c2baa0c", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "Not dependent on a prior visual moment.", "_shot_index": 1, "visible_entities": [ "L06", "C01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 6, "shot_index": 1 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At torso height beside the bicycle, the camera tracks parallel at a stable medium lateral distance, preserving 수리영's energetic profile and the circular rhythm of her pedaling. She occupies the middle-left with one bent knee driving the pedal downward, while her bicycle, trailing hair, road, and coastal context extend toward the right without changing the established side axis.", "lighting_mood": "Daylight is treated with a cool desaturated palette, restrained contrast, and crisp tactile detail in the moving hair and clothing.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "movement_direction", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "midground", "gesture_action": "moves_toward", "gesture_target_label": "road ahead", "label": "수리영 riding the bicycle", "screen_zone": "middle_left", "target_id": "C01", "target_kind": "character" } ] }, "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": [ "L06", "C01" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "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" ] }, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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" } ], "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" ] }, "render_contracts": [], "_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": "d65436aab138959a" }, { "scene_index": 7, "heading": "S#7. 유기견 보호원 / D", "beat_title": "사납게 짖어대는 로트와일러", "representative_moment": "C11이 철창 안에서 이빨을 드러낸 채 정면을 향해 앞발을 치켜들고 있다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Close eye-level frontal confrontation, cool desaturated daylight", "t2i_prompt": "Photorealistic cinematic still. C11O00 stands in the middle-center midground, facing the camera behind enclosure bars occupying the middle-center foreground. Positioned at the animal's eye level, C11O00 shows traits carried from stable_traits — a huge size, black fur, a thick snout, and a heavy head. The dog lunges forward with one forepaw raised and teeth bared in a deliberate frontal confrontation, gaze fixed directly on the camera. [L07: Close vertical metal bars form a solid barrier right in front]. Daylight appropriate to the shelter is rendered in cool desaturated tones, casting a restrained low-key contrast across the dark form.", "outfit_assignments": [ { "character_id": "C11", "outlook_id": "O00" } ], "source_facts": [ "S#7 action: '사납게 짖으며 날뛰는 로트와일러'", "shot description: '철창 안에서 이빨을 드러낸 채 앞발을 치켜든 거대한 검은 개의 정면'" ], "visual_inferences": [ "fixed_element shelter_daylight_lighting: cool desaturated tones and low-key contrast applied to the dark fur", "The dog confronts the viewer through metal bars, amplifying the threatening tension" ], "creative_decisions": [ "Eye-level perspective to make the confrontation direct and intimidating", "Using the enclosure bars as foreground elements to create spatial depth while keeping the subject completely visible" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002" ], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "083807e53330c660", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "69afcd385190040f", "owned_usage_hash": "4f53cda18c2baa0c", "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": [ "L07", "C11" ], "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": "At the dog's eye level and close to the enclosure, the static camera centers 거대한 검은 개 in a deliberate frontal confrontation while keeping the bars visible across the foreground. Its raised forepaw and bared teeth are contained within a close frame rather than cropped into an abstract detail, making the threat read through posture as well as expression.", "lighting_mood": "Daylight appropriate to the shelter is rendered in cool desaturated tones with restrained low-key contrast across the dog's dark form.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "primary_subject_isolation", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "foreground", "gesture_action": "none", "gesture_target_label": "", "label": "enclosure bars", "screen_zone": "middle_center", "target_id": "", "target_kind": "background" }, { "constraint_id": "fsc_002", "depth_plane": "midground", "gesture_action": "none", "gesture_target_label": "", "label": "거대한 검은 개", "screen_zone": "middle_center", "target_id": "C11", "target_kind": "character" } ] }, "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": [ "L07", "C11" ], "allowed_outlook_pairs": [ { "character_id": "C11", "outlook_id": "O00" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "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", "describe the close subject and its immediate surroundings without relying on the absent background reference", "use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "shelter_daylight_lighting", "element_type": "environment_state", "character_name": "", "description": "The interior of the animal shelter is illuminated by daylight characterized by cool, desaturated tones. The lighting maintains a restrained, low-key contrast, creating a subdued and consistent atmospheric tone across the facility space.", "applies_to_shots": [ 1, 5 ], "element_scope": "full", "source_facts": [ "Daylight appropriate to the shelter is rendered in cool desaturated tones with restrained low-key contrast across the dog's dark form (Shot 1).", "Daylight appropriate to the shelter is kept cool and desaturated, with restrained contrast separating the scissors, fingers, and neck fur (Shot 5)." ], "visual_inferences": [ "The light source is natural daylight filtering into the utilitarian shelter, maintaining a uniform cool color temperature throughout the room." ], "creative_decisions": [ "Using a desaturated palette and low-key lighting to emphasize the calm yet somewhat sterile and bleak atmosphere of the animal shelter." ], "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [], "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" ] }, "render_contracts": [], "_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": "34c1dd00144c56a3" }, { "scene_index": 7, "heading": "S#7. 유기견 보호원 / D", "beat_title": "바쁘게 돌아가는 유기견 보호원 업무", "representative_moment": "An insert of C01's hands carefully trimming the thick neck fur of C11 with scissors.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "hand-height insert composition with shallow depth of field, cool desaturated daylight", "t2i_prompt": "Photorealistic cinematic still. An insert composition at hand height. C01 in a casual hooded jacket, an Asian young woman, with her supporting hand entering from the lower side and her dominant hand holding silver scissors, meticulously trimming the fur. The scissor blades cross the center diagonally. At the opposite frame edge, a small section of C11, a giant black dog, its head angled away, with its thick neck and black fur remaining visible as a natural scale reference. The background is completely soft and out-of-focus. The interior of the animal shelter is illuminated by cool, desaturated daylight, maintaining a restrained, low-key contrast that sharply separates the metallic scissors, bare fingers, and dark neck fur.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C11", "outlook_id": "O00" } ], "source_facts": [ "S7 action: '달려가 털깎고, 연고 바르고, 목욕시키고등'", "Shot description: 개의 목덜미 털을 가위로 조심스럽게 잘라내는 수리영의 손" ], "visual_inferences": [ "fixed_element shelter_daylight_lighting: cool, desaturated daylight with restrained low-key contrast.", "As an insert framing scale, the focus is entirely isolated on the hands, scissors, and a small section of the dog's fur, omitting the full bodies." ], "creative_decisions": [ "Positioned the scissor blades diagonally across the center to create a dynamic intersection between the human hands and the animal's neck.", "Utilized a soft, out-of-focus background to enforce the insert framing rule and isolate the tactile texture of the fur and metallic tools." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "180307bade208c05", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "02c3abb5c11a1d6c", "owned_usage_hash": "4f53cda18c2baa0c", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 7, "dependency_reason": "Continuity of lighting and spatial environment (shelter_daylight_lighting) established in Scene 7, Shot 1.", "_shot_index": 5, "visible_entities": [ "C01", "C11" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 7, "shot_index": 5 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "insert", "moment_lock": "single still moment from the selected shot only", "camera_direction": "After the cut, the camera sits at hand height beside 수리영 and isolates her hands, the scissors, and a small section of the dog's neck within an insert composition. The scissor blades cross the center diagonally while her supporting hand enters from the lower side and the dog's neck remains at the opposite edge as a natural scale reference.", "lighting_mood": "Daylight appropriate to the shelter is kept cool and desaturated, with restrained contrast separating the scissors, fingers, and neck fur.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": null, "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", "C11" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C11", "outlook_id": "O00" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [ { "subject_id": "C01", "policy_type": "identity_reference", "policy": "base_id_required", "reason": "The insert shows 수리영 primarily through her hands, so a full outlook reference is not visually alignable." }, { "subject_id": "C11", "policy_type": "identity_reference", "policy": "base_id_required", "reason": "Only a limited portion of the dog's neck and head edge is visible in the grooming insert." } ], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "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" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "shelter_daylight_lighting", "element_type": "environment_state", "character_name": "", "description": "The interior of the animal shelter is illuminated by daylight characterized by cool, desaturated tones. The lighting maintains a restrained, low-key contrast, creating a subdued and consistent atmospheric tone across the facility space.", "applies_to_shots": [ 1, 5 ], "element_scope": "full", "source_facts": [ "Daylight appropriate to the shelter is rendered in cool desaturated tones with restrained low-key contrast across the dog's dark form (Shot 1).", "Daylight appropriate to the shelter is kept cool and desaturated, with restrained contrast separating the scissors, fingers, and neck fur (Shot 5)." ], "visual_inferences": [ "The light source is natural daylight filtering into the utilitarian shelter, maintaining a uniform cool color temperature throughout the room." ], "creative_decisions": [ "Using a desaturated palette and low-key lighting to emphasize the calm yet somewhat sterile and bleak atmosphere of the animal shelter." ], "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character", "id": "C01", "policy": "required" }, { "kind": "character", "id": "C11", "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" ] }, "render_contracts": [], "_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": "304cd39d0ce9c387" }, { "scene_index": 8, "heading": "번화가·대형 마트 / 해질 무렵", "beat_title": "수리영의 번화가 이동", "representative_moment": "C01 steering her bicycle through the busy neon-lit alley at dusk.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "waist-height tracking shot with cool desaturated dusk lighting", "t2i_prompt": "Photorealistic cinematic still. C01O01 in a casual hooded jumper, a young Korean woman, steering her bicycle, occupies the lower-center midground at a front three-quarter angle from the right, moving toward the neon-lined alley route. Seen from a waist-height observer, her torso and arms are in frame, hands turning the handlebar, with the bicycle frame and front wheel visible below. She is mid-pedal, weight pressing one pedal in the down stroke, her dark hair flowing back in the direction opposite to motion, looking off-screen. Behind her, illuminated neon signs face the camera, establishing the commercial alley. The road stretches backward, framed by the already present cars parked at oblique angles and street stalls opening toward the sidewalk. Dusk ambient light creates a cool desaturated mood with restrained contrast, while selective deep-crimson and gold accents from the streetlights and storefronts trace the environment.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "수리영 rides a bicycle through the busy street.", "She and the bicycle occupy the lower-center in a front three-quarter view." ], "visual_inferences": [ "fixed_element suyeong_base_appearance: casual and practical outfit suitable for riding a bicycle in the evening.", "The action of turning the handlebar and loading one pedal is frozen as mid-pedal with hair flowing back." ], "creative_decisions": [ "Positioned the camera at waist height to emphasize the bicycle motion while keeping the background layers visible.", "Applied a cool desaturated palette with crimson and gold accents to match the visual world rules and lighting mood." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001" ], "reference_phrase_kinds": [ "character", "background" ], "owned_object_usage": [ { "owned_token": "AC unit", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "awning", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "bollard", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "building", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "car", "usage_kind": "anchor", "source_phrase": "already present cars parked at oblique angles" }, { "owned_token": "crate", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "flowerpot", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "gas meter", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "gas pipe", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "laundry line", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "power line", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "road", "usage_kind": "anchor", "source_phrase": "The road stretches backward" }, { "owned_token": "security grille", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "shopping cart", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "sidewalk", "usage_kind": "anchor", "source_phrase": "street stalls opening toward the sidewalk" }, { "owned_token": "sign", "usage_kind": "anchor", "source_phrase": "illuminated neon signs face the camera" }, { "owned_token": "storefront", "usage_kind": "anchor", "source_phrase": "streetlights and storefronts" }, { "owned_token": "streetlight", "usage_kind": "anchor", "source_phrase": "streetlights and storefronts" }, { "owned_token": "supermarket", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "truck", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "utility pole", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "0da84a1f3302f562", "owned_hash": "ea86161bfeb678fd", "camera_direction_hash": "79656b556e347f40", "owned_usage_hash": "a393581374c0fd71", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "car", "violating_phrase": "already present cars parked at oblique angles", "reason": "The prompt refers to already present cars in the background as a spatial anchor, which is consistent with the producer's anchor classification.", "verdict": "anchor_reference" }, { "owned_object": "road", "violating_phrase": "The road stretches backward", "reason": "The prompt refers to the road to describe the spatial layout of the background, aligning with anchor usage.", "verdict": "anchor_reference" }, { "owned_object": "sidewalk", "violating_phrase": "street stalls opening toward the sidewalk", "reason": "The prompt mentions the sidewalk purely as a reference point for the orientation of street stalls, confirming it as an anchor.", "verdict": "anchor_reference" }, { "owned_object": "sign", "violating_phrase": "illuminated neon signs face the camera", "reason": "The prompt refers to existing background neon signs to establish the scene, consistent with anchor usage.", "verdict": "anchor_reference" }, { "owned_object": "storefront", "violating_phrase": "streetlights and storefronts", "reason": "The prompt mentions storefronts only as a source of background light accents, which is consistent with anchor usage.", "verdict": "anchor_reference" }, { "owned_object": "streetlight", "violating_phrase": "streetlights and storefronts", "reason": "The prompt mentions streetlights only as a source of selective environmental lighting accents, confirming its anchor role.", "verdict": "anchor_reference" } ] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "", "_shot_index": 1, "visible_entities": [ "C01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 8, "shot_index": 1 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "wide", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At waist height about 2.5 meters ahead and offset toward 수리영's front side, the camera tracks backward and briefly steadies on the instant she turns the handlebar and loads one pedal. She and the bicycle occupy the lower-center in a front three-quarter view, while the neon signs, taxis, street stalls, and alley activity remain distributed behind her as readable but secondary layers.", "lighting_mood": "Dusk and the illuminated neon signs support a cool desaturated street image with restrained contrast and selective deep-crimson and gold accents.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "movement_direction", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "midground", "gesture_action": "moves_toward", "gesture_target_label": "neon-lined alley route", "label": "수리영 steering her bicycle", "screen_zone": "lower_center", "target_id": "C01", "target_kind": "character" } ] }, "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" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L09B01", "reference_usage": "exact_background", "owned_objects": [ "AC unit", "awning", "bollard", "building", "car", "crate", "door", "flowerpot", "gas meter", "gas pipe", "laundry line", "power line", "road", "security grille", "shopping cart", "sidewalk", "sign", "storefront", "streetlight", "supermarket", "truck", "utility pole", "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "suyeong_base_appearance", "element_type": "character_state", "character_name": "수리영", "description": "A young Korean woman with consistent facial features and hairstyle. She wears the same casual and practical outfit suitable for riding a bicycle in the evening. Her overall physical appearance and clothing remain entirely unchanged across the locations.", "applies_to_shots": [ 1, 2 ], "element_scope": "full", "source_facts": [ "수리영 rides a bicycle through the busy street.", "She parks her bicycle and immediately enters the mart to find her mother." ], "visual_inferences": [ "Because she transitions directly from the street to the mart, her clothing and hair remain exactly the same.", "Her outfit is practical for cycling in an urban environment at dusk." ], "creative_decisions": [ "Selected a casual daily outfit appropriate for a young woman living in a port town to maintain a realistic atmosphere." ], "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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": "background", "id": "L09B01", "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" ] }, "render_contracts": [], "_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": "8d09cb205ac64dc4" }, { "scene_index": 8, "heading": "S#8. 번화가·대형 마트 / 해질 무렵", "beat_title": "엇갈린 단서와 멈칫하는 수리영", "representative_moment": "마트 통로에서 직원으로부터 엄마의 행방을 전해 들은 수리영(C01)이 혼란스러운 표정으로 고개를 갸웃하며 정지해 있는 순간.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "At Suyeong's eye level, medium two-shot angled across the shared eyeline", "t2i_prompt": "Photorealistic cinematic still. At eye level, a medium two-shot captures two figures separated by a shared supermarket aisle space. In the left foreground, C01O01 in a casual hooded jumper, a 21-year-old Korean woman with an oval face and tied black hair, stands at a three-quarter angle from the right. Her head is halted in a slight tilt, eyes looking off-screen in confusion. Across the shared aisle in the right midground, a middle-aged Korean female employee in worker uniform stands at a three-quarter angle from the left, leaning slightly toward the customer, her eyes fixed on C01O01. Behind them, the aisle-facing sides of the shelf recede diagonally, creating depth. The scene is illuminated by ambient supermarket ceiling light at dusk, rendered in a cool desaturated palette with restrained low-key contrast.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "수리영(C01) arrives at the mart and looks for her mother.", "The female employee informs Suyeong that her mother left early.", "Suyeong is confused by the unexpected news.", "Shot description: observer at Suyeong's eye level, medium two-shot." ], "visual_inferences": [ "fixed_element suyeong_base_appearance: Suyeong's casual cycling outfit is preserved from the previous shot.", "The unregistered female employee wears a typical supermarket worker uniform.", "Shelf lines and ambient lighting define the supermarket interior." ], "creative_decisions": [ "Placed Suyeong in the left foreground and the employee in the right midground to emphasize their spatial relationship and interaction.", "Used ambient ceiling light with a cool desaturated palette to highlight the dusk setting and growing tension." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [], "reference_phrase_kinds": [ "character", "background" ], "owned_object_usage": [ { "owned_token": "basket", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "card terminal", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "cardboard box", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "ceiling light", "usage_kind": "anchor", "source_phrase": "ambient supermarket ceiling light at dusk" }, { "owned_token": "checkout counter", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "price label", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "produce island", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "refrigerator case", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "restocking cart", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "security camera", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "service counter", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "shelf", "usage_kind": "anchor", "source_phrase": "aisle-facing sides of the shelf recede diagonally" }, { "owned_token": "shopping cart", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "sliding door", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "143ea60337959fda", "owned_hash": "6aa50095efec2848", "camera_direction_hash": "d35ed18ff3c2b186", "owned_usage_hash": "057190131b23db10", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "shelf", "violating_phrase": "the aisle-facing sides of the shelf recede diagonally", "reason": "The shelf is mentioned solely as a compositional anchor to establish depth behind the characters, matching the producer's anchor declaration.", "verdict": "anchor_reference" }, { "owned_object": "ceiling light", "violating_phrase": "ambient supermarket ceiling light", "reason": "The ceiling light is referenced only as a light source to define the ambient illumination and mood, matching the producer's anchor declaration.", "verdict": "anchor_reference" } ] } } ], "scene_type": "normal", "dependent_scene_index": 8, "dependency_reason": "exact_background rule: shares the same supermarket space (L09) as the previous shot, maintaining Suyeong's appearance and the established room.", "_shot_index": 2, "visible_entities": [ "L09", "C01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 8, "shot_index": 2 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At 수리영’s eye level, the dolly-in settles into a medium two-shot angled across the shared eyeline, with 수리영 in the left foreground and the 한국인 여성 직원 offset in the right midground. 수리영’s halted head tilt becomes the only changed axis, while shelf lines behind them preserve the inward trajectory and leave lateral room for the subsequent pan.", "lighting_mood": "Ambient supermarket lighting at dusk is rendered in a cool desaturated palette with restrained low-key contrast.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": null, "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", "C01" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L09B02", "reference_usage": "exact_background", "owned_objects": [ "basket", "card terminal", "cardboard box", "ceiling light", "checkout counter", "price label", "produce island", "refrigerator case", "restocking cart", "security camera", "service counter", "shelf", "shopping cart", "sliding 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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "suyeong_base_appearance", "element_type": "character_state", "character_name": "수리영", "description": "A young Korean woman with consistent facial features and hairstyle. She wears the same casual and practical outfit suitable for riding a bicycle in the evening. Her overall physical appearance and clothing remain entirely unchanged across the locations.", "applies_to_shots": [ 1, 2 ], "element_scope": "full", "source_facts": [ "수리영 rides a bicycle through the busy street.", "She parks her bicycle and immediately enters the mart to find her mother." ], "visual_inferences": [ "Because she transitions directly from the street to the mart, her clothing and hair remain exactly the same.", "Her outfit is practical for cycling in an urban environment at dusk." ], "creative_decisions": [ "Selected a casual daily outfit appropriate for a young woman living in a port town to maintain a realistic atmosphere." ], "confidence": "medium" } ], "previous_shot_refs": [ { "scene_index": 8, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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": "background", "id": "L09B02", "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" ] }, "render_contracts": [], "_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": "7c3843eaeb9d110a" }, { "scene_index": 9, "heading": "S#9. 해안가 / N", "beat_title": "어둠 속 질주", "representative_moment": "C01이 어두운 해안도로 위에서 상체를 굽힌 채 자전거 페달을 밟으며 질주하는 측면 실루엣.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "tightened strict side profile tracking parallel from rider-torso height", "t2i_prompt": "Photorealistic cinematic still. C01 in a dark hooded jacket, an Asian young woman in her 20s, occupies the middle-center midground, moving toward the road ahead. Captured in strict left profile from rider-torso height tracking parallel, she is bent low over the handlebar, driving the pedal fully down in a mid-pedal freeze. The bicycle's frame and wheels are visible beneath her, anchored to the lateral movement line of the coastal road. Her long tied hair and jacket flap trail backward in the direction opposite to her forward motion. Night ambient light renders her as a cool desaturated side silhouette with restrained contrast, her body volume preserved against the dim background while her face remains swallowed in deep shadow. Looking off-screen, her compressed diagonal posture fills the medium frame, sparse streetlights casting faint illumination over the empty asphalt.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "S#9 action: '해안도로를 자전거를 타고 바삐 집으로 달려가는 수리영.'", "shot description: 어두워진 해안도로 위, 상체를 잔뜩 굽힌 채 한쪽 페달을 아래로 딛은 찰나의 수리영의 측면 실루엣, 옷자락이 뒤로 강하게 휘날리는", "camera_direction: From rider-torso height, tracking parallel in a tightened strict side profile" ], "visual_inferences": [ "Night ambient light and cool desaturated palette to enhance the mysterious silhouette", "Sparse streetlights on the coastal road to provide spatial scale without breaking the silhouette", "Bicycle frame and wheels included to ground the pedaling action physically" ], "creative_decisions": [ "Used a mid-pedal freeze to convey horizontal speed while adhering to the single-moment rule", "Described hair and clothes trailing opposite to motion to emphasize momentum", "Obscured facial details in deep shadow to align with the silhouette requirement while preserving body volume depth" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001" ], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "dd6f3eead274a3e1", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "4e23cc25b03a7806", "owned_usage_hash": "4f53cda18c2baa0c", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 6, "dependency_reason": "이전 씬(S6)에서 등장한 해안도로의 공간적 분위기(atmosphere_reference)와 어둠의 톤을 연속해서 유지", "_shot_index": 1, "visible_entities": [ "L06", "C01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 9, "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 rider-torso height, the camera tracks parallel in a tightened strict side profile, keeping 수리영 and the bicycle moving horizontally through the middle of the frame. Her bent torso and the pedal at its lowest point form a compressed diagonal, while enough road remains visible around the silhouette to retain speed and scale.", "lighting_mood": "Night ambient light renders the rider as a cool desaturated side silhouette with restrained contrast.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "movement_direction", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "midground", "gesture_action": "moves_toward", "gesture_target_label": "road ahead", "label": "수리영", "screen_zone": "middle_center", "target_id": "C01", "target_kind": "character" } ] }, "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": [ "L06", "C01" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [ { "subject_id": "C01", "policy_type": "identity_reference", "policy": "base_id_required", "reason": "수리영 is defined primarily as a dark side silhouette in motion, so an outlook reference would add unsupported visible detail." } ], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "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" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [ { "scene_index": 6, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character", "id": "C01", "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" ] }, "render_contracts": [], "_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": "73d36e16860ff04e" }, { "scene_index": 10, "heading": "S#10. 옥탑방 / N", "beat_title": "미세하게 열린 현관문", "representative_moment": "잠금장치가 풀려 아주 살짝 틈이 벌어진 어두운 옥탑방 철문.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Waist-height close framing, restrained low-key night lighting", "t2i_prompt": "Photorealistic cinematic still. The narrow opening of rooftop room door stands in the middle-center midground. [L03: Observed directly from outside at waist height, the heavy iron door rests slightly ajar, its outer face angled slightly away from the camera. The worn metal door frame outlines the separation, exposing a thin vertical recess of absolute darkness inside. The surrounding exterior walls display aging textures lit by a cool, desaturated night ambient light, maintaining a restrained low-key contrast that keeps the interior starkly shadowed and hidden.]", "outfit_assignments": [], "source_facts": [ "S#10. 옥탑방 / N", "문이 미세하게 열리고", "Shot description: released iron door slightly ajar exposing narrow dark interior" ], "visual_inferences": [ "fixed_element rooftop_exterior_night_environment: cool, dark, and desaturated ambient lighting against worn building textures", "waist-height camera capturing the vertical slice of darkness to emphasize the threshold" ], "creative_decisions": [ "Focusing purely on the slightly opened door gap to build claustrophobic suspense without revealing the attacker or victim", "Applying low-key contrast to highlight the rusted textures of the door and frame against the pitch-black interior" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001" ], "reference_phrase_kinds": [], "owned_object_usage": [ { "owned_token": "AC", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "clothesline", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "curtain", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "flowerpot", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "gas meter", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "house", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "iron door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "pipe", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "railing", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "rooftop", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "rooftop room", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "utility pole", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "water tank", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wire", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "08c07294ada97723", "owned_hash": "bb1fb263d6853c1b", "camera_direction_hash": "94fd0dc60642ce21", "owned_usage_hash": "be4460c704d73abc", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 4, "dependency_reason": "옥탑방 외부 배경 텍스처 및 조명톤 참조", "_shot_index": 3, "visible_entities": [ "L03" ], "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": "Directly outside the doorway at waist height, the retreat finishes and briefly locks on the narrow separation between the released iron door and its frame. The door occupies the central midground without exceeding a natural portion of the image, while the dark interior is visible only as a thin vertical recess.", "lighting_mood": "Night ambient light leaves the doorway in restrained low-key contrast, with the interior held dark.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "required_background_position", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "midground", "gesture_action": "none", "gesture_target_label": "", "label": "narrow opening of rooftop room door", "screen_zone": "middle_center", "target_id": "", "target_kind": "background" } ] }, "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": [ "L03" ], "allowed_outlook_pairs": [], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "L03B03", "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", "describe the close subject and its immediate surroundings without relying on the absent background reference", "use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "rooftop_exterior_night_environment", "element_type": "environment_state", "character_name": "", "description": "The nighttime exterior environment of an old rooftop room on a multi-family villa. The surrounding ambient lighting is cool, dark, and desaturated, creating a low-key contrast against the worn, aging textures of the building exterior walls. The atmosphere is quiet and starkly isolated in the darkness.", "applies_to_shots": [ 3, 4 ], "element_scope": "full", "source_facts": [ "S#10. 옥탑방 / N", "그 시각. 옥탑방 옥상.", "Night ambient light leaves the doorway in restrained low-key contrast", "cool, desaturated night exterior" ], "visual_inferences": [ "The building facade shows aging and wear, consistent with an old rooftop structure." ], "creative_decisions": [ "Maintaining a consistent cool, desaturated color palette for the exterior to emphasize the thriller atmosphere and highlight any localized practical lights." ], "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [], "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" ] }, "render_contracts": [], "_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": "0b0ecbede994788c" }, { "scene_index": 10, "heading": "S#10. 옥탑방 / N", "beat_title": "Shadow of the assault on the lit window", "representative_moment": "The overlapping shadows of C08 violently lunging over the bent C03 are cast against the glowing rooftop room window.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Chest-height frontal medium shot, stark silhouette contrast against glowing warm window, surrounded by cool desaturated night exterior", "t2i_prompt": "Photorealistic cinematic still. A lit rooftop room window in the middle-center midground glows against the dark exterior, revealing stark projections where C08 in a dark long coat, an Asian adult man, appearing as a shadow in the middle-center background, looks toward C03 in a home wear, a Korean middle-aged woman, appearing as a shadow in the lower-center background. C08, the figure exerting force, is captured in three-quarter angle from the left, lunging forcefully and pinning the other down from above. C03, the figure receiving the force, is in right profile, folding abruptly beneath the attacker in a severely injured posture, her head looking down. The decisive, overlapping dark silhouettes flatten the body volume into sharp outlines against the illuminated curtain. The camera sits at chest height in a stable frontal alignment, retaining the window boundary symmetrically around the dramatic assault, isolated within the cool, desaturated night environment.", "outfit_assignments": [ { "character_id": "C03", "outlook_id": "O03" }, { "character_id": "C08", "outlook_id": "O08" } ], "source_facts": [ "S#10 action: 남자가 안으로 들어가 창문에 비친 그림자. 민숙을 덮쳐 어깨를 물어뜯는 듯한 모습.", "shot description: frontal surface shows the large shadow of the man overlapping 강민숙's bent shadow" ], "visual_inferences": [ "fixed_element rooftop_exterior_night_environment applied as cool desaturated night exterior", "shadow projection flattens body volume against the illuminated window curtain", "attacker exerts force pinning the victim down from above in a dramatic silhouette" ], "creative_decisions": [ "Chest-height frontal alignment to lock the window squarely in the frame.", "Stark silhouette contrast to emphasize the violence strictly through the sharp shadow shapes." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002", "fsc_003" ], "reference_phrase_kinds": [ "character", "background" ], "owned_object_usage": [ { "owned_token": "AC", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "clothesline", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "curtain", "usage_kind": "anchor", "source_phrase": "illuminated curtain" }, { "owned_token": "flowerpot", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "gas meter", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "house", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "iron door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "pipe", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "railing", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "rooftop", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "rooftop room", "usage_kind": "anchor", "source_phrase": "lit rooftop room window" }, { "owned_token": "utility pole", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "water tank", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "anchor", "source_phrase": "lit rooftop room window" }, { "owned_token": "wire", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "1802324593da59c9", "owned_hash": "bb1fb263d6853c1b", "camera_direction_hash": "37e566ab40cc05ac", "owned_usage_hash": "1c962180fa4e4f3e", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "rooftop room", "violating_phrase": "lit rooftop room window", "reason": "The phrase references the existing rooftop room window as a spatial anchor/descriptor, which matches the producer's 'anchor' declaration.", "verdict": "anchor_reference" }, { "owned_object": "window", "violating_phrase": "lit rooftop room window", "reason": "The window is referenced as a compositional frame and backdrop for the shadow projections, which matches the producer's 'anchor' declaration.", "verdict": "anchor_reference" }, { "owned_object": "curtain", "violating_phrase": "illuminated curtain", "reason": "The curtain is mentioned as an illuminated background to show the silhouettes, acting as a compositional anchor and matching the producer's 'anchor' declaration.", "verdict": "anchor_reference" } ] } } ], "scene_type": "normal", "dependent_scene_index": 10, "dependency_reason": "Background and spatial layout must maintain exact consistency with the established rooftop exterior in Scene 10, Shot 3.", "_shot_index": 4, "visible_entities": [ "L03", "C03", "C08" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 10, "shot_index": 4 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "The lateral track ends at chest height in a stable frontal alignment with the lit window, placing the overlapping shadows in the center while retaining the window boundary around them. The 낯선 한국인 남성’s enlarged upper shadow descends from above and behind 강민숙’s bent shadow, pinning her silhouette into the lower portion of the pane.", "lighting_mood": "The lit window isolates the figures as stark silhouettes within an otherwise cool, desaturated night exterior.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "shared_space_relation", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "midground", "gesture_action": "none", "gesture_target_label": "", "label": "lit rooftop room window", "screen_zone": "middle_center", "target_id": "", "target_kind": "background" }, { "constraint_id": "fsc_002", "depth_plane": "background", "gesture_action": "none", "gesture_target_label": "", "label": "강민숙 shadow", "screen_zone": "lower_center", "target_id": "C03", "target_kind": "character" }, { "constraint_id": "fsc_003", "depth_plane": "background", "gesture_action": "looks_toward", "gesture_target_label": "강민숙's shadow", "label": "낯선 한국인 남성 shadow", "screen_zone": "middle_center", "target_id": "C08", "target_kind": "character" } ] }, "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": [ "L03", "C03", "C08" ], "allowed_outlook_pairs": [ { "character_id": "C03", "outlook_id": "O03" }, { "character_id": "C08", "outlook_id": "O08" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [ { "subject_id": "C03", "policy_type": "identity_reference", "policy": "base_id_required", "reason": "강민숙 appears solely as a bent shadow behind the window surface." }, { "subject_id": "C08", "policy_type": "identity_reference", "policy": "base_id_required", "reason": "Only the characters’ overlapping shadows are visible, so outlook-specific visual references should not introduce facial or wardrobe detail." } ], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L03B03", "reference_usage": "exact_background", "owned_objects": [ "AC", "clothesline", "curtain", "flowerpot", "gas meter", "house", "iron door", "pipe", "railing", "rooftop", "rooftop room", "utility pole", "water tank", "window", "wire" ], "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "rooftop_exterior_night_environment", "element_type": "environment_state", "character_name": "", "description": "The nighttime exterior environment of an old rooftop room on a multi-family villa. The surrounding ambient lighting is cool, dark, and desaturated, creating a low-key contrast against the worn, aging textures of the building exterior walls. The atmosphere is quiet and starkly isolated in the darkness.", "applies_to_shots": [ 3, 4 ], "element_scope": "full", "source_facts": [ "S#10. 옥탑방 / N", "그 시각. 옥탑방 옥상.", "Night ambient light leaves the doorway in restrained low-key contrast", "cool, desaturated night exterior" ], "visual_inferences": [ "The building facade shows aging and wear, consistent with an old rooftop structure." ], "creative_decisions": [ "Maintaining a consistent cool, desaturated color palette for the exterior to emphasize the thriller atmosphere and highlight any localized practical lights." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 10, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character", "id": "C03", "policy": "required" }, { "kind": "character", "id": "C08", "policy": "required" }, { "kind": "background", "id": "L03B03", "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" ] }, "render_contracts": [], "_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": "24b259199ca3a80c" }, { "scene_index": 11, "heading": "S#11. 옥탑 방 밖 / N", "beat_title": "역겨운 냄새와 반쯤 열린 현관문", "representative_moment": "코를 가린 수리영이 불안한 눈빛으로 반쯤 열린 옥탑방 철문을 응시한다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "face-dominant close framing at eye level, cool desaturated night lighting", "t2i_prompt": "Photorealistic cinematic still. C01O01 in a casual hooded jumper, a 21-year-old Korean woman with long black hair, occupies the middle-left foreground in three-quarter left profile, looking toward the [: narrow strip of a half-open iron door] in the middle-right background. The camera is positioned at eye level in a face-dominant close framing, focusing tightly on her anxious expression as one hand is clamped firmly over her nose. Her tense off-screen gaze is fixed strictly on the door edge. The immediate rooftop surroundings are enveloped in cool, desaturated natural night ambience, with restrained low-key contrast casting deep shadows across the side of her face and the metallic surface at the frame edge.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "S#11 action: 수리영 feels a repulsive smell on the rooftop.", "shot description: hand clamped over nose, gazing anxiously at the half-open iron door." ], "visual_inferences": [ "Close framing isolates her face and hand to emphasize the sensory shock and suspense.", "[L02: half-open iron door] is partially visible at the frame edge to serve as the spatial anchor for her gaze without relying on a full background reference." ], "creative_decisions": [ "Applied 'middle-left foreground' and 'middle-right background' mapping per frame spatial contract.", "Used cool, desaturated tones and low-key contrast to heighten the eerie, claustrophobic occult thriller tone." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002" ], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [ { "owned_token": "AC", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "clothesline", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "drying rack", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "flowerpot", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "landing", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "meter", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "neighboring house", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "pipe", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "railing", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "rooftop room", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "security bar", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "stair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "utility pole", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "water tank", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wire", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "489a7b86c20a3901", "owned_hash": "7b12e19738bbaefa", "camera_direction_hash": "798f1795f0cd52a0", "owned_usage_hash": "8e60d83a88732a0a", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "Not dependent on a prior shot for continuous background, utilizing skipped_close_framing mode.", "_shot_index": 3, "visible_entities": [ "C01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 11, "shot_index": 3 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At 수리영’s eye level and just opposite the half-open door, the dolly-in ends in a face-dominant close framing with her hand still covering her nose. She occupies the near side of the frame in three-quarter view, and a narrow strip of the half-open iron door remains at the opposite edge to establish the direction of her anxious gaze.", "lighting_mood": "Natural night ambience appropriate to the rooftop is kept cool and desaturated with restrained low-key contrast.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "looks_to_anchor", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "background", "gesture_action": "none", "gesture_target_label": "", "label": "half-open iron door", "screen_zone": "middle_right", "target_id": "", "target_kind": "background" }, { "constraint_id": "fsc_002", "depth_plane": "foreground", "gesture_action": "looks_toward", "gesture_target_label": "half-open iron door", "label": "수리영", "screen_zone": "middle_left", "target_id": "C01", "target_kind": "character" } ] }, "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" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "L04B02", "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", "describe the close subject and its immediate surroundings without relying on the absent background reference", "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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" } ], "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" ] }, "render_contracts": [], "_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": "6bcd2fcf6c154946" }, { "scene_index": 12, "heading": "S#12. 옥탑방 안 / N", "beat_title": "민숙의 참혹한 죽음 발견", "representative_moment": "C03가 커튼 뒤에 목과 어깨가 뜯긴 채 죽어 앉아 있다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Observer POV close framing at shoulder height, cool desaturated low-key lighting with sharp crimson contrast", "t2i_prompt": "Photorealistic cinematic still. C03O03 in comfortable home wear, a Korean woman in her early 40s, sits lifeless in the middle-center midground. Seen at three-quarter angle from the right from a shoulder-height observer perspective, the figure is slumped with head lowered, looking down. The head, neck, and upper chest occupy most of the center and lower frame, exposing stark deep-crimson torn wounds across the shoulder and collarbone with thick blood trailing down the skin. A displaced fabric curtain edge forms a narrow vertical border in the foreground beside the figure. [L05: faint ambient stand light casts cool desaturated low-key shadows around the immediate floor area, highlighting a pool of deep-crimson blood beneath the seated body].", "outfit_assignments": [ { "character_id": "C03", "outlook_id": "O03" } ], "source_facts": [ "shot description: tight view of 강민숙 seated behind the displaced curtain, keeping her lowered head and the torn neck, shoulder, and collarbone wounds together.", "scene text: 앉은 채로 고개를 떨군 채 죽어있는 엄마- 민숙!!!! 살점이 뜯긴 어깨와 쇄골" ], "visual_inferences": [ "fixed_element rooftop_room_lighting: faint, desaturated, and cool low-key lighting originating from a stand light, deep-crimson accents", "immobilized C03: torso slumped and head dropped forward, seated behind the curtain" ], "creative_decisions": [ "Used a close framing scale with the curtain edge as a foreground element to create depth without obscuring the grisly wounds", "Emphasized the stark color contrast between the cool shadows and the vivid deep-crimson blood to elevate the horror mood" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001" ], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [ { "owned_token": "bed", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "bedding", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "bloodstain", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "ceiling light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "curtain", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "dining table", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "floor", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wall", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wall marking", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "1452aff29dd2059a", "owned_hash": "f212e7ed366b4268", "camera_direction_hash": "32890be01d035f22", "owned_usage_hash": "92f4031a499f113b", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 5, "dependency_reason": "rooftop_room_lighting continuous from previous scene location context", "_shot_index": 3, "visible_entities": [ "L05", "C03" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 12, "shot_index": 3 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From just behind and beside 수리영’s prior position at shoulder height, the push-in stops on a tight view of 강민숙 seated behind the displaced curtain, keeping her lowered head and the torn neck, shoulder, and collarbone wounds together. Her body occupies most of the center and lower frame, while the curtain edge remains as a narrow foreground border rather than obscuring the injuries.", "lighting_mood": "The room’s established faint stand light supports cool desaturated low-key illumination, with blood and exposed wounds retained as deep-crimson accents.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "primary_subject_isolation", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "midground", "gesture_action": "none", "gesture_target_label": "", "label": "강민숙", "screen_zone": "middle_center", "target_id": "C03", "target_kind": "character" } ] }, "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": [ "L05", "C03" ], "allowed_outlook_pairs": [ { "character_id": "C03", "outlook_id": "O03" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "L05B05", "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", "describe the close subject and its immediate surroundings without relying on the absent background reference", "use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "rooftop_room_lighting", "element_type": "environment_state", "character_name": "", "description": "The interior is bathed in faint, desaturated, and cool low-key lighting originating from a stand light. This subdued and dim atmosphere serves as a consistent backdrop throughout the space. Against this dark environment, specific focal points appear as stark, deep-crimson accents, creating a vivid color contrast.", "applies_to_shots": [ 3, 8, 9 ], "element_scope": "full", "source_facts": [ "The room’s established faint stand light supports cool desaturated low-key illumination", "blood and exposed wounds retained as deep-crimson accents", "Faint established room light keeps the image cool and subdued", "transformed tattoo reads as a concentrated deep-crimson accent", "established faint room illumination remains cool and low-key", "isolating the red circle as the sole saturated accent" ], "visual_inferences": [ "The ambient lighting remains universally dim and blue-tinted across all camera angles, allowing only the red elements to pop visually." ], "creative_decisions": [ "Using stark color contrast (cool, desaturated shadows versus highly saturated deep crimson) to visually unify the eerie and shocking sequence." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 5, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [], "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" ] }, "render_contracts": [], "_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": "1c70f44eea8bbb58" }, { "scene_index": 12, "heading": "S#12. 옥탑방 안 / N /수리영 방", "beat_title": "붉게 변하는 문신과 기괴한 현상의 시작", "representative_moment": "An extreme close-up of C03's lifeless wrist, revealing a tattoo that has vividly turned red beside a crumpled photograph.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Extreme close insert at hand height, dolly-in, highlighting a concentrated deep-crimson accent against faint cool low-key lighting.", "t2i_prompt": "Photorealistic cinematic still. C03 in a comfortable homewear sleeve, a Korean woman in her 40s, with her marked wrist occupying the middle-center foreground. In this isolated extreme close insert, the lifeless wrist crosses the frame diagonally. The marked side of the wrist faces the camera, displaying a tattoo that has distinctly changed to a concentrated deep-crimson red. The resting hand lies motionless next to the partial edge of a crumpled photograph, its pictured content unreadable in this tight framing. The background is a soft, out-of-focus void, bathed in faint, desaturated, and cool low-key lighting originating from an off-screen stand light, isolating the vivid red mark as the sole saturated accent.", "outfit_assignments": [ { "character_id": "C03", "outlook_id": "O03" } ], "source_facts": [ "Shot description: extreme close view of the wrist at the instant the tattoo becomes distinctly red.", "Shot description: only the edge of the crumpled photograph and a small area of the resting hand remain for narrative context.", "S#12 text: 엄마 손목에 있던 문신이 붉은색으로 변한다." ], "visual_inferences": [ "fixed_element rooftop_room_lighting: The faint established room light keeps the image cool and subdued.", "The hand is lifeless and motionless, reflecting the immobilized dead state of the character.", "The tattoo reads as a highly saturated deep-crimson accent against the dim environment." ], "creative_decisions": [ "Utilized an isolated extreme close framing (insert) to exclude the full body and concentrate entirely on the supernatural color shift of the tattoo.", "Rendered the background as an out-of-focus void to strictly adhere to the single primary-subject framing rule for close-ups." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001" ], "reference_phrase_kinds": [ "character", "background" ], "owned_object_usage": [ { "owned_token": "bed", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "bedding", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "bloodstain", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "ceiling light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "curtain", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "dining table", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "floor", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wall", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wall marking", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "b3cabec25505e152", "owned_hash": "f212e7ed366b4268", "camera_direction_hash": "3c844c642240a402", "owned_usage_hash": "92f4031a499f113b", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 5, "dependency_reason": "exact_background (maintaining the established room lighting and environment from prior rooftop room scenes)", "_shot_index": 8, "visible_entities": [ "C03" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 12, "shot_index": 8 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "insert", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At hand height beside 강민숙, the dolly-in tightens into an isolated extreme close view of the wrist at the instant the tattoo becomes distinctly red. The wrist crosses the center diagonally, while only the edge of the crumpled photograph and a small area of the resting hand remain for narrative context.", "lighting_mood": "Faint established room light keeps the image cool and subdued while the transformed tattoo reads as a concentrated deep-crimson accent.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "required_background_position", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "foreground", "gesture_action": "none", "gesture_target_label": "", "label": "강민숙's marked wrist", "screen_zone": "middle_center", "target_id": "C03", "target_kind": "character" } ] }, "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" ], "allowed_outlook_pairs": [ { "character_id": "C03", "outlook_id": "O03" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [ { "subject_id": "C03", "policy_type": "identity_reference", "policy": "base_id_required", "reason": "Only 강민숙’s wrist and part of her hand are visible, making a full outlook reference unsuitable." } ], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L05B05", "reference_usage": "exact_background", "owned_objects": [ "bed", "bedding", "bloodstain", "ceiling light", "chair", "curtain", "dining table", "door", "floor", "wall", "wall marking", "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "rooftop_room_lighting", "element_type": "environment_state", "character_name": "", "description": "The interior is bathed in faint, desaturated, and cool low-key lighting originating from a stand light. This subdued and dim atmosphere serves as a consistent backdrop throughout the space. Against this dark environment, specific focal points appear as stark, deep-crimson accents, creating a vivid color contrast.", "applies_to_shots": [ 3, 8, 9 ], "element_scope": "full", "source_facts": [ "The room’s established faint stand light supports cool desaturated low-key illumination", "blood and exposed wounds retained as deep-crimson accents", "Faint established room light keeps the image cool and subdued", "transformed tattoo reads as a concentrated deep-crimson accent", "established faint room illumination remains cool and low-key", "isolating the red circle as the sole saturated accent" ], "visual_inferences": [ "The ambient lighting remains universally dim and blue-tinted across all camera angles, allowing only the red elements to pop visually." ], "creative_decisions": [ "Using stark color contrast (cool, desaturated shadows versus highly saturated deep crimson) to visually unify the eerie and shocking sequence." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 5, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character", "id": "C03", "policy": "required" }, { "kind": "background", "id": "L05B05", "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" ] }, "render_contracts": [], "_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": "182677c07562035c" }, { "scene_index": 12, "heading": "S#12. 옥탑방 안 / N", "beat_title": "기괴한 초자연적 현상 발생", "representative_moment": "침대 위쪽 벽면에 선명한 붉은 원(P02)이 나타나며 방안의 차가운 조명과 강렬하게 대비되는 순간.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Medium framing from a bedside position with a frontal tilt, focusing entirely on the wall and the mark", "t2i_prompt": "Photorealistic cinematic still. P02, the red circle on the wall above the bed, is centered in the middle-center background, while the upper part of the bed rests in the lower-center midground. The vertical surface of the wall is presented nearly frontally. The interior is bathed in faint, desaturated, and cool low-key lighting originating from a stand light. This subdued and dim atmosphere serves as a consistent backdrop, isolating the stark, deep-crimson pigment of P02 as the sole saturated visual accent in the room.", "outfit_assignments": [], "source_facts": [ "shot description: 침대 위쪽 벽면에 붉은색 원 형태가 선명하게 그려진 모습", "S#12 action: 침대 위 벽에 붉은 원이 생겨난다" ], "visual_inferences": [ "fixed_element rooftop_room_lighting: The ambient lighting remains universally dim and blue-tinted, allowing only the red elements to pop visually", "The camera completes its rising tilt to settle frontally on the wall, using the bed purely as a spatial anchor at the bottom edge" ], "creative_decisions": [ "Applied medium framing to isolate the supernatural P02 mark on the wall while maintaining the bed edge for realistic spatial context", "Used intense deep-crimson saturation for P02 to create a stark color contrast against the cool, desaturated background" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002" ], "reference_phrase_kinds": [ "background" ], "owned_object_usage": [ { "owned_token": "bed", "usage_kind": "anchor", "source_phrase": "the upper part of the bed" }, { "owned_token": "bedding", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "bloodstain", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "ceiling light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "curtain", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "dining table", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "floor", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wall", "usage_kind": "anchor", "source_phrase": "the wall above the bed" }, { "owned_token": "wall marking", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "c9b5ececd62be811", "owned_hash": "f212e7ed366b4268", "camera_direction_hash": "d202ba54897b62a5", "owned_usage_hash": "8f8e56c457a23b1c", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "bed", "violating_phrase": "the upper part of the bed", "reason": "The bed is used purely as a spatial and compositional reference, which aligns with the producer's anchor echo.", "verdict": "anchor_reference" }, { "owned_object": "wall", "violating_phrase": "on the wall above the bed", "reason": "The wall is mentioned only to anchor the spatial layout and establish the frontal composition, matching the producer's anchor echo.", "verdict": "anchor_reference" }, { "owned_object": "wall marking", "violating_phrase": "P02, the red circle on the wall", "reason": "The red circle on the wall represents a change in the visual state of the wall marking, which is explicitly and structurally required by the shot intent ('침대 위쪽 벽면에 선명한 붉은 원(P02)이 나타나며').", "verdict": "allowed_visual_state_change" } ] } } ], "scene_type": "normal", "dependent_scene_index": 12, "dependency_reason": "S12_Shot3 배경 참조 (동일한 옥탑방 공간 유지)", "_shot_index": 9, "visible_entities": [ "L05", "P02" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 12, "shot_index": 9 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From the same bedside position, the camera completes its rising tilt and settles on the wall above the bed with the red circle centered. The upper portion of the bed remains along the lower edge as a restrained spatial reference, while the marked wall occupies the primary field without becoming an abstract insert.", "lighting_mood": "The established faint room illumination remains cool and low-key, isolating the red circle as the sole saturated accent.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "required_background_position", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "background", "gesture_action": "none", "gesture_target_label": "", "label": "red circle on wall above bed", "screen_zone": "middle_center", "target_id": "", "target_kind": "background" }, { "constraint_id": "fsc_002", "depth_plane": "midground", "gesture_action": "none", "gesture_target_label": "", "label": "upper part of bed", "screen_zone": "lower_center", "target_id": "", "target_kind": "background" } ] }, "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": [ "L05", "P02" ], "allowed_outlook_pairs": [], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L05B05", "reference_usage": "exact_background", "owned_objects": [ "bed", "bedding", "bloodstain", "ceiling light", "chair", "curtain", "dining table", "door", "floor", "wall", "wall marking", "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "rooftop_room_lighting", "element_type": "environment_state", "character_name": "", "description": "The interior is bathed in faint, desaturated, and cool low-key lighting originating from a stand light. This subdued and dim atmosphere serves as a consistent backdrop throughout the space. Against this dark environment, specific focal points appear as stark, deep-crimson accents, creating a vivid color contrast.", "applies_to_shots": [ 3, 8, 9 ], "element_scope": "full", "source_facts": [ "The room’s established faint stand light supports cool desaturated low-key illumination", "blood and exposed wounds retained as deep-crimson accents", "Faint established room light keeps the image cool and subdued", "transformed tattoo reads as a concentrated deep-crimson accent", "established faint room illumination remains cool and low-key", "isolating the red circle as the sole saturated accent" ], "visual_inferences": [ "The ambient lighting remains universally dim and blue-tinted across all camera angles, allowing only the red elements to pop visually." ], "creative_decisions": [ "Using stark color contrast (cool, desaturated shadows versus highly saturated deep crimson) to visually unify the eerie and shocking sequence." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 12, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "background", "id": "L05B05", "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" ] }, "render_contracts": [], "_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": "e724d2d46346cfc5" }, { "scene_index": 13, "heading": "S#13. 옥탑방 외곽 / N", "beat_title": "수리영의 충격과 혜수의 긴급한 접근", "representative_moment": "C02 runs mid-stride toward C01, who stands blankly in front of the flashing police car.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "wide full-body composition, eye-level tracking backward, cool desaturated night palette with restrained contrast", "t2i_prompt": "Photorealistic cinematic still. C01O01 in a casual hoodie jumper, a 21-year-old Korean woman, stands in the middle-right background, while C02O02 in a leather jacket, a middle-aged Korean woman, in the middle-left midground moves toward C01O01. At eye height in a wide full-body composition, C01O01 is seen at three-quarter angle from the left, slack with weight unshifted, gazing into the distance. C02O02 is captured at three-quarter angle from the right, running mid-stride with one foot lifted off the shared courtyard they both occupy, eyes fixed on C01O01. Flashing lights from the police car behind C01O01 punctuate a cool desaturated night palette with restrained contrast, grounding both women. The lower approach of the exterior stair stands visibly behind the open space.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C02", "outlook_id": "O02" } ], "source_facts": [ "S#13 action: '차에서 내린 혜수... 초점을 잃고 멍하니 서있는 수리영의 모습을 발견하고 달려간다'", "shot description: 경광등이 번쩍이는 경찰차 앞 초점을 잃고 멍하니 선 수리영과, 그를 향해 달려오다 한 발이 땅에서 떨어진 mid-action 순간의 혜수의 전신" ], "visual_inferences": [ "cool desaturated night atmosphere from the scene time/mood", "flashing police lights hitting the courtyard and characters" ], "creative_decisions": [ "used wide framing to capture both full bodies with foreground/background depth and a shared spatial anchor (the courtyard)", "mid-stride pose to emphasize Hye-su's urgency while maintaining physical consistency", "anchored the background with the police car and exterior stair directly from the scene description" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002" ], "reference_phrase_kinds": [ "character", "background" ], "owned_object_usage": [ { "owned_token": "AC unit", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "awning", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "bicycle", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "clothesline", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "courtyard", "usage_kind": "anchor", "source_phrase": "shared courtyard they both occupy" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "drainpipe", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "exterior stair", "usage_kind": "anchor", "source_phrase": "exterior stair stands visibly" }, { "owned_token": "flowerpot", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "gas meter", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "gas pipe", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "gate", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "police car", "usage_kind": "anchor", "source_phrase": "police car behind C01O01" }, { "owned_token": "power line", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "security bar", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "utility pole", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "villa", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "water tank", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "5f2be46c52f9ded8", "owned_hash": "5e646253a44c38d2", "camera_direction_hash": "388c072ea9f4c31f", "owned_usage_hash": "d5877bb726d88403", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "courtyard", "violating_phrase": "shared courtyard they both occupy", "reason": "The prompt references the courtyard solely as a spatial ground for the characters' positions, which is a valid anchor reference.", "verdict": "anchor_reference" }, { "owned_object": "exterior stair", "violating_phrase": "exterior stair stands visibly", "reason": "The prompt references the exterior stair solely to describe its presence and position in the background, which is a valid anchor reference.", "verdict": "anchor_reference" }, { "owned_object": "police car", "violating_phrase": "Flashing lights from the police car", "reason": "The prompt describes the flashing lights of the police car, which is a visual state change explicitly required by the shot intent ('flashing police car'), satisfying exception (1).", "verdict": "allowed_visual_state_change" } ] } } ], "scene_type": "normal", "dependent_scene_index": 11, "dependency_reason": "shares the exact background location (L04 courtyard and exterior stairs) established in previous scenes", "_shot_index": 2, "visible_entities": [ "L04", "C01", "C02" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 13, "shot_index": 2 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "wide", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At eye height below the stairs, the camera tracks backward along 혜수’s approach line in a wide full-body composition, holding 수리영 near the police car in the background and 혜수 in the advancing midground. The frame is timed to 혜수’s lifted foot, while 수리영 remains fixed and unfocused so only their relative positions change.", "lighting_mood": "Flashing police-car light punctuates a cool desaturated night palette while restrained contrast keeps both women grounded and readable.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "movement_direction", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "background", "gesture_action": "none", "gesture_target_label": "", "label": "수리영", "screen_zone": "middle_right", "target_id": "C01", "target_kind": "character" }, { "constraint_id": "fsc_002", "depth_plane": "midground", "gesture_action": "moves_toward", "gesture_target_label": "수리영", "label": "혜수", "screen_zone": "middle_left", "target_id": "C02", "target_kind": "character" } ] }, "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", "C01", "C02" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C02", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L04B01", "reference_usage": "exact_background", "owned_objects": [ "AC unit", "awning", "bicycle", "clothesline", "courtyard", "door", "drainpipe", "exterior stair", "flowerpot", "gas meter", "gas pipe", "gate", "police car", "power line", "security bar", "utility pole", "villa", "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [ { "scene_index": 11, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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": "C02O02", "policy": "required" }, { "kind": "background", "id": "L04B01", "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" ] }, "render_contracts": [], "_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": "d3f0269b3336186e" }, { "scene_index": 14, "heading": "S#14. 옥탑방 안 / N", "beat_title": "사라진 흔적과 깨끗해진 방", "representative_moment": "수리영의 시선으로 바라본, 핏자국 하나 없이 비정상적으로 깨끗하게 정돈된 침대와 텅 빈 커튼 뒤 공간", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "standing eye-level wide perspective, cool desaturated night ambient lighting", "t2i_prompt": "Photorealistic cinematic still. Seen from a standing eye-level perspective at the threshold, the bed area without red mark sits meticulously organized in the middle-left midground, while the empty space behind curtain is fully revealed in the middle-right background. The interior space is illuminated by a restrained ambient night light, casting a cool desaturated tone across the unnaturally clean environment. The bed remains neatly offset on one side, with the blank wall above it showing absolutely no traces of blood or supernatural markings. On the other side, the opened curtain exposes the thoroughly vacant floor, emphasizing the complete absence of any physical evidence in the single coherent layout.", "outfit_assignments": [], "source_facts": [ "지나칠 정도로 깨끗하게 정돈된 방안. 핏자국도 보이지 않고", "침대 위에 붉은 표식은 물론 커튼 뒤의 시신도 감쪽 같이 사라졌다.", "The bed remains offset on one side while the opened curtain and vacant floor occupy the other", "the meticulously ordered, empty room remains legible beyond her fixed eyeline" ], "visual_inferences": [ "fixed_element sterile_empty_bedroom: interior is meticulously organized and unnaturally clean, showing no traces of blood or a body", "fixed_element cool_night_interior_lighting: restrained ambient night light with a cool, desaturated color palette" ], "creative_decisions": [ "Wide eye-level perspective from the threshold to establish the entire sterile space, allowing both the bed and the area behind the curtain to be seen simultaneously, emphasizing the uncanny emptiness." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002" ], "reference_phrase_kinds": [ "background" ], "owned_object_usage": [ { "owned_token": "bed", "usage_kind": "anchor", "source_phrase": "the bed area without red mark sits meticulously organized" }, { "owned_token": "ceiling light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "curtain", "usage_kind": "anchor", "source_phrase": "the empty space behind curtain is fully revealed" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "table", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "417aeb09ade74d4f", "owned_hash": "553a983bfd5399fa", "camera_direction_hash": "9b44bdb58b4e4ba3", "owned_usage_hash": "8ec6df69b19112f7", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "bed", "violating_phrase": "the bed area without red mark sits meticulously organized", "reason": "The prompt describes the bed's existing clean state and layout position as a composition anchor, which is consistent with the producer's declaration of anchor.", "verdict": "anchor_reference" }, { "owned_object": "curtain", "violating_phrase": "the empty space behind curtain is fully revealed", "reason": "The prompt refers to the curtain to define the scene's spatial layout and composition without instructing it to be redrawn, which is consistent with the producer's declaration of anchor.", "verdict": "anchor_reference" } ] } } ], "scene_type": "normal", "dependent_scene_index": 12, "dependency_reason": "수리영의 방 안 공간 배경과 조명 분위기를 이전 씬(시신이 있었던 때)과 동일한 구조로 유지하되, 내용물(시신, 핏자국, 표식)만 사라진 상태를 대비시키기 위해 이전 샷을 exact_background 참조", "_shot_index": 3, "visible_entities": [ "L05" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 14, "shot_index": 3 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "wide", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From standing eye level on the threshold, a wide pan sweeps across the bed area and completes at the cleared curtain, then holds on the empty space behind it. The bed remains offset on one side while the opened curtain and vacant floor occupy the other, allowing the excessive order and total absence of blood, body, or wall mark to read in one coherent layout.", "lighting_mood": "Interior ambient light is treated in a cool desaturated palette with restrained contrast, emphasizing the room’s unnaturally complete visual order.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "required_background_position", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "midground", "gesture_action": "none", "gesture_target_label": "", "label": "bed area without red mark", "screen_zone": "middle_left", "target_id": "", "target_kind": "background" }, { "constraint_id": "fsc_002", "depth_plane": "background", "gesture_action": "none", "gesture_target_label": "", "label": "empty space behind curtain", "screen_zone": "middle_right", "target_id": "", "target_kind": "background" } ] }, "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": [ "L05" ], "allowed_outlook_pairs": [], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L05B06", "reference_usage": "exact_background", "owned_objects": [ "bed", "ceiling light", "chair", "curtain", "door", "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "sterile_empty_bedroom", "element_type": "environment_state", "character_name": "", "description": "The rooftop bedroom interior is meticulously organized and unnaturally clean, showing absolutely no traces of blood, markings, or a body. The bed remains neatly positioned on one side, while the space behind the opened curtain and the adjacent floor are completely vacant and orderly.", "applies_to_shots": [ 3, 4 ], "element_scope": "full", "source_facts": [ "지나칠 정도로 깨끗하게 정돈된 방안. 핏자국도 보이지 않고", "침대 위에 붉은 표식은 물론 커튼 뒤의 시신도 감쪽 같이 사라졌다.", "The bed remains offset on one side while the opened curtain and vacant floor occupy the other", "the meticulously ordered, empty room remains legible beyond her fixed eyeline" ], "visual_inferences": [ "The complete absence of the crime scene is visually emphasized through extreme tidiness, making the space feel sterile to contrast with the character's panic." ], "creative_decisions": [ "Maintaining a deep focus or wide layout that allows the excessive order and the empty space behind the curtain to read clearly as a single, undisturbed environment." ], "confidence": "high" }, { "element_id": "cool_night_interior_lighting", "element_type": "environment_state", "character_name": "", "description": "The interior spaces of the apartment are illuminated by a restrained ambient night light. The overall lighting maintains a consistently cool, desaturated color palette with low contrast, casting a subdued and somewhat sterile visual tone across the environment.", "applies_to_shots": [ 3, 4, 8 ], "element_scope": "full", "source_facts": [ "Interior ambient light is treated in a cool desaturated palette with restrained contrast", "Restrained ambient night interior light supports a cool, desaturated palette", "Neutral ambient interior light is shaped with restrained contrast and a cool, desaturated finish" ], "visual_inferences": [], "creative_decisions": [ "Applying a uniform cool and desaturated lighting atmosphere across both the bedroom and the dining area to visually reinforce the psychological chill and the unnerving emptiness of the location." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 12, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "background", "id": "L05B06", "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" ] }, "render_contracts": [], "_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": "eb3f20361d0020e1" }, { "scene_index": 14, "heading": "S#14. 옥탑방 안 / N", "beat_title": "사라진 시신과 핏자국", "representative_moment": "C01 rushes into her room only to find it unnervingly clean, her pale face staring at the empty space.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "observer POV close composition with shallow focus, cool desaturated night palette", "t2i_prompt": "Photorealistic cinematic still. C01O01 in a casual hooded jumper, a 21-year-old Korean woman, occupies the middle-left foreground, her pale oval face at three-quarter angle from the right, looking off-screen toward the empty room interior in the middle-right background. Her shoulders are locked and she leans forward slightly, her long black hair framing her features. The camera holds a close composition with shallow focus, isolating her bewildered expression while [: the meticulously organized, unnaturally clean rooftop bedroom] remains softly legible beyond her eyeline. The long side of the neatly positioned bed recedes behind her, its upper surface visible, leading to an opened curtain and a completely vacant floor in the negative space. Restrained ambient night interior light bathes the environment in a cool, desaturated palette with low contrast, keeping the sterile background subdued behind her pallor.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "지나칠 정도로 깨끗하게 정돈된 방안. 핏자국도 보이지 않고", "침대 위에 붉은 표식은 물론 커튼 뒤의 시신도 감쪽 같이 사라졌다.", "얼굴이 파리해지는 수리영.", "The bed remains offset on one side while the opened curtain and vacant floor occupy the other", "the meticulously ordered, empty room remains legible beyond her fixed eyeline" ], "visual_inferences": [ "fixed_element sterile_empty_bedroom: bed neatly positioned, vacant floor, complete absence of the crime scene", "fixed_element cool_night_interior_lighting: restrained ambient light, low contrast", "Soo-ri-young's locked shoulders and leaning posture reflect her sudden shock at the missing body" ], "creative_decisions": [ "Shallow focus close framing on Soo-ri-young's three-quarter profile in the middle-left foreground, with her eyeline directing the viewer to the softly resolved empty room in the middle-right background", "Applying a cool, desaturated night palette to contrast her pale skin with the unnervingly sterile environment", "Replaced the background location tag L05 with L01 to comply with the modification request" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002" ], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [ { "owned_token": "bed", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "ceiling light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "curtain", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "table", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "d61e8a02ab3c735b", "owned_hash": "553a983bfd5399fa", "camera_direction_hash": "d76269e124591161", "owned_usage_hash": "7b3a0d4ba1cdc16d", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 14, "dependency_reason": "Shares the identical apartment setting and narrative timeline as the previous shot, demanding strict environmental and lighting continuity.", "_shot_index": 4, "visible_entities": [ "C01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 14, "shot_index": 4 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From inside the room at 수리영's eye level, finish the dolly-in on her front three-quarter side, letting her pale face occupy most of the left-center frame while the meticulously ordered, empty room remains legible beyond her fixed eyeline. Hold the close composition just before moving alongside her, with shallow focus preserving the absence behind her rather than isolating her as a portrait.", "lighting_mood": "Restrained ambient night interior light supports a cool, desaturated palette and keeps the empty room subdued behind 수리영's pallor.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "looks_to_anchor", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "background", "gesture_action": "none", "gesture_target_label": "", "label": "empty room interior", "screen_zone": "middle_right", "target_id": "", "target_kind": "background" }, { "constraint_id": "fsc_002", "depth_plane": "foreground", "gesture_action": "looks_toward", "gesture_target_label": "empty room interior", "label": "수리영", "screen_zone": "middle_left", "target_id": "C01", "target_kind": "character" } ] }, "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" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "L05B06", "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", "describe the close subject and its immediate surroundings without relying on the absent background reference", "use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "sterile_empty_bedroom", "element_type": "environment_state", "character_name": "", "description": "The rooftop bedroom interior is meticulously organized and unnaturally clean, showing absolutely no traces of blood, markings, or a body. The bed remains neatly positioned on one side, while the space behind the opened curtain and the adjacent floor are completely vacant and orderly.", "applies_to_shots": [ 3, 4 ], "element_scope": "full", "source_facts": [ "지나칠 정도로 깨끗하게 정돈된 방안. 핏자국도 보이지 않고", "침대 위에 붉은 표식은 물론 커튼 뒤의 시신도 감쪽 같이 사라졌다.", "The bed remains offset on one side while the opened curtain and vacant floor occupy the other", "the meticulously ordered, empty room remains legible beyond her fixed eyeline" ], "visual_inferences": [ "The complete absence of the crime scene is visually emphasized through extreme tidiness, making the space feel sterile to contrast with the character's panic." ], "creative_decisions": [ "Maintaining a deep focus or wide layout that allows the excessive order and the empty space behind the curtain to read clearly as a single, undisturbed environment." ], "confidence": "high" }, { "element_id": "cool_night_interior_lighting", "element_type": "environment_state", "character_name": "", "description": "The interior spaces of the apartment are illuminated by a restrained ambient night light. The overall lighting maintains a consistently cool, desaturated color palette with low contrast, casting a subdued and somewhat sterile visual tone across the environment.", "applies_to_shots": [ 3, 4, 8 ], "element_scope": "full", "source_facts": [ "Interior ambient light is treated in a cool desaturated palette with restrained contrast", "Restrained ambient night interior light supports a cool, desaturated palette", "Neutral ambient interior light is shaped with restrained contrast and a cool, desaturated finish" ], "visual_inferences": [], "creative_decisions": [ "Applying a uniform cool and desaturated lighting atmosphere across both the bedroom and the dining area to visually reinforce the psychological chill and the unnerving emptiness of the location." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 14, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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" } ], "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" ] }, "render_contracts": [], "_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": "67b81f9a496ae4cf" }, { "scene_index": 14, "heading": "S#14. 옥탑방 안 / N", "beat_title": "사라진 시신과 혜수의 위로", "representative_moment": "혜수가 수리영의 어깨를 두 손으로 부드럽게 감싸 쥐며 다독인다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "medium framing from upper-torso height, cool desaturated lighting with low contrast", "t2i_prompt": "Photorealistic cinematic still. Shot from upper-torso height. C02O02 in a leather jacket, a Korean middle-aged woman, occupies the center frame from the waist up, positioned at a three-quarter angle from the left. Her hands gently clasp the shoulders of C01O01 in a casual hooded jumper, a Korean young woman. C01O01 is situated in the lower foreground with her back partially to the camera, both characters anchored in the same room. C01O01 has her shoulders drawn inward, gaze directed down. C02O02's eyes are fixed on C01O01 with a reassuring expression. The muted background features an orderly window and door. Neutral ambient interior light shapes the scene with restrained contrast and a cool, desaturated finish, keeping visual attention on the hands and tense shoulders.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C02", "outlook_id": "O02" } ], "source_facts": [ "수리영의 어깨를 두 손으로 부드럽게 감싸 쥔 혜수의 상체", "지나칠 정도로 깨끗하게 정돈된 방안", "혜수가 혼란스러워하는 수리영을 감싸며 다독임" ], "visual_inferences": [ "fixed_element cool_night_interior_lighting: The interior spaces of the apartment are illuminated by a restrained ambient night light.", "C01's shoulders drawn inward defensively, reflecting shock and confusion.", "C02 leaning in with a protective posture." ], "creative_decisions": [ "Medium framing anchored at upper-torso height to emphasize the physical contact of the hands on the shoulders.", "Cool, desaturated finish applied to contrast the eerie cleanliness of the room with the emotional warmth of the comforting gesture." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [], "reference_phrase_kinds": [ "character", "background" ], "owned_object_usage": [ { "owned_token": "bed", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "ceiling light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "curtain", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "anchor", "source_phrase": "and door" }, { "owned_token": "table", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "anchor", "source_phrase": "orderly window" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "ff4377140bebc1de", "owned_hash": "553a983bfd5399fa", "camera_direction_hash": "598be750e9e5d8bb", "owned_usage_hash": "57e35d7578506c97", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "door", "violating_phrase": "orderly window and door", "reason": "The door is described as a passive background element to anchor the scene, matching the producer's anchor declaration with no redraw intent.", "verdict": "anchor_reference" }, { "owned_object": "window", "violating_phrase": "orderly window", "reason": "The window is mentioned as a passive background feature to establish context, matching the producer's anchor declaration with no redraw intent.", "verdict": "anchor_reference" } ] } } ], "scene_type": "normal", "dependent_scene_index": 14, "dependency_reason": "Same room and time as Scene 14 Shot 4. Background and lighting must remain consistent.", "_shot_index": 8, "visible_entities": [ "C01", "C02" ], "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": "At upper-torso height and one arm's length away, complete the gentle dolly-in toward the shared axis, framing 혜수 from the upper torso while both hands close around 수리영's shoulders in the lower third. Keep 수리영 partially back-to-camera so the physical reassurance, rather than either face alone, becomes the compositional center before the downward tilt begins.", "lighting_mood": "Neutral ambient interior light is shaped with restrained contrast and a cool, desaturated finish, keeping attention on the hands and tense shoulders.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": null, "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" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C02", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L05B06", "reference_usage": "exact_background", "owned_objects": [ "bed", "ceiling light", "chair", "curtain", "door", "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "cool_night_interior_lighting", "element_type": "environment_state", "character_name": "", "description": "The interior spaces of the apartment are illuminated by a restrained ambient night light. The overall lighting maintains a consistently cool, desaturated color palette with low contrast, casting a subdued and somewhat sterile visual tone across the environment.", "applies_to_shots": [ 3, 4, 8 ], "element_scope": "full", "source_facts": [ "Interior ambient light is treated in a cool desaturated palette with restrained contrast", "Restrained ambient night interior light supports a cool, desaturated palette", "Neutral ambient interior light is shaped with restrained contrast and a cool, desaturated finish" ], "visual_inferences": [], "creative_decisions": [ "Applying a uniform cool and desaturated lighting atmosphere across both the bedroom and the dining area to visually reinforce the psychological chill and the unnerving emptiness of the location." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 14, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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": "C02O02", "policy": "required" }, { "kind": "background", "id": "L05B06", "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" ] }, "render_contracts": [], "_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": "dcdb966d1428f960" }, { "scene_index": 15, "heading": "S#15. 해안가. 버스정류장 부근 / N", "beat_title": "멍하니 바다를 응시하는 수리영", "representative_moment": "C01이 어두운 밤바다 앞 벤치에 홀로 멍하니 앉아 있다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "static wide view from seated eye level, cool desaturated low-key lighting", "t2i_prompt": "Photorealistic cinematic still. C01O01 in a casual hooded jumper, a young Korean woman with an oval face and tied long black straight hair, sits slumped on the bench edge in the middle-left midground, looking toward the dark sea in the middle-right background. Captured from a seated eye-level wide view, the figure is seen in right profile, gazing into the distance across the broad negative space. The empty bus stop bench runs obliquely across the frame, exposing its front and one receding side. A coastal bus stop structure frames the isolated seating, one side facing the camera while its length recedes toward the water. Restrained low-key lighting casts deep shadows around her stillness, while a cool, desaturated palette dominates the coastal night environment, emphasizing the sparse atmospheric setting.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "S#15. 해안가. 버스정류장 부근 / N", "바다를 바라보며 멍한 채로 앉아있는 수리영", "Dark coastal night ambience is rendered in a cool, desaturated palette with restrained low-key contrast." ], "visual_inferences": [ "The lighting remains consistently dim, utilizing shadows to emphasize the isolation of the location.", "C01 slumped on bench edge in right profile, gazing into the distance." ], "creative_decisions": [ "Placed C01 off-center in the middle-left to create broad negative space toward the sea.", "Oblique bench angle to naturalize the distant eyeline without crossing the lens." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002" ], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "d3b482f5d6c52327", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "6f6dc0665556b7ec", "owned_usage_hash": "4f53cda18c2baa0c", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "", "_shot_index": 1, "visible_entities": [ "L11", "C01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 15, "shot_index": 1 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "wide", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Hold a static wide view from seated eye level in front of the bench, placing 수리영 off-center while the night sea remains visible behind and to the side of her. Angle the bench and her body so her distant eyeline can travel naturally toward the sea without crossing into the lens, leaving broad negative space around her stillness.", "lighting_mood": "Dark coastal night ambience is rendered in a cool, desaturated palette with restrained low-key contrast.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "looks_to_anchor", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "background", "gesture_action": "none", "gesture_target_label": "", "label": "sea", "screen_zone": "middle_right", "target_id": "", "target_kind": "background" }, { "constraint_id": "fsc_002", "depth_plane": "midground", "gesture_action": "looks_toward", "gesture_target_label": "sea", "label": "수리영", "screen_zone": "middle_left", "target_id": "C01", "target_kind": "character" } ] }, "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": [ "L11", "C01" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "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" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "coastal_night_environment", "element_type": "environment_state", "character_name": "", "description": "The scene is set in a dark coastal area at night, characterized by a cool, desaturated color palette and restrained low-key lighting. The ambient darkness of the seaside dominates the background, establishing a sparse and atmospheric outdoor setting.", "applies_to_shots": [ 1, 5, 6 ], "element_scope": "full", "source_facts": [ "S#15. 해안가. 버스정류장 부근 / N", "Dark coastal night ambience is rendered in a cool, desaturated palette with restrained low-key contrast." ], "visual_inferences": [ "The lighting remains consistently dim, utilizing shadows to emphasize the isolation of the location." ], "creative_decisions": [ "Applying a cool color grade across all shots to unify the visual tone and heighten the desolate, tense mood of the scene." ], "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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" } ], "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" ] }, "render_contracts": [], "_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": "92022fe520f0d843" }, { "scene_index": 15, "heading": "S#15. 해안가. 버스정류장 부근 / N", "beat_title": "수리영의 도주와 혜수의 외침", "representative_moment": "해안가 도로에서 C02가 양손을 입가에 모으고 달려가는 C01의 등을 향해 소리친다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "chest-height tracking medium shot, cool desaturated night palette", "t2i_prompt": "Photorealistic cinematic still. C02O02 in a casual leather jacket, a Korean middle-aged woman, stands in the middle-left foreground looking toward the departing C01O01. C01O01 in a casual hood jumper, a young Korean woman, appears in the upper-right background moving toward the departure path, both sharing the same coastal road. Framed at chest height beside C02O02, her torso and arms occupy the foreground with her lower body implied beyond the bottom frame edge, captured at three-quarter angle from the right. Her hands are cupped firmly around her mouth projecting her voice, her mature oval face showing a tense expression, eyes fixed on C01O01. Positioned diagonally beyond C02O02's shoulder, C01O01 is captured small with her back to camera, caught mid-run away, her limbs in an active running posture, long black straight hair flowing in the direction opposite to her motion. The dark coastal environment features a cool, desaturated color palette and restrained low-key lighting, maintaining a dimly lit peripheral sea that emphasizes the distance between the two figures.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C02", "outlook_id": "O02" } ], "source_facts": [ "수리영: back_to_camera, caught mid-run away, gaze=off_screen", "혜수: hands cupped around mouth, gaze=looks_at_character→C01", "S#15. 해안가. 버스정류장 부근 / N" ], "visual_inferences": [ "fixed_element hye_soo_shouting: upper body is engaged in the act of shouting", "fixed_element suri_young_running: limbs are in an active running motion", "fixed_element coastal_night_environment: lighting remains consistently dim" ], "creative_decisions": [ "Applied a chest-height frame constraint, making Hye-soo's torso dominant while her lower body is implied beyond the bottom edge.", "Used a cool desaturated palette to highlight the spatial tension between the foreground caller and the receding background figure." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002" ], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "1f45774ac3c8e0e8", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "79a962ba8ef668eb", "owned_usage_hash": "4f53cda18c2baa0c", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 15, "dependency_reason": "이전 샷과 동일한 해안가 야간 환경이며, 공간적 연속성과 조명 분위기(cool, desaturated night ambience)를 유지함", "_shot_index": 5, "visible_entities": [ "L11", "C01", "C02" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 15, "shot_index": 5 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Track beside 혜수 at chest height and tighten to her upper body, holding her in the left foreground as her cupped hands reach her mouth. Keep 수리영 small and receding along the same diagonal beyond 혜수's shoulder, so the departing direction remains unmistakable without widening away from 혜수's call.", "lighting_mood": "Available night ambience remains cool and desaturated, with restrained contrast separating 혜수 from the receding figure.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "movement_direction", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "background", "gesture_action": "moves_toward", "gesture_target_label": "departure path", "label": "수리영", "screen_zone": "upper_right", "target_id": "C01", "target_kind": "character" }, { "constraint_id": "fsc_002", "depth_plane": "foreground", "gesture_action": "looks_toward", "gesture_target_label": "departing 수리영", "label": "혜수", "screen_zone": "middle_left", "target_id": "C02", "target_kind": "character" } ] }, "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": [ "L11", "C01", "C02" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C02", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "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" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "coastal_night_environment", "element_type": "environment_state", "character_name": "", "description": "The scene is set in a dark coastal area at night, characterized by a cool, desaturated color palette and restrained low-key lighting. The ambient darkness of the seaside dominates the background, establishing a sparse and atmospheric outdoor setting.", "applies_to_shots": [ 1, 5, 6 ], "element_scope": "full", "source_facts": [ "S#15. 해안가. 버스정류장 부근 / N", "Dark coastal night ambience is rendered in a cool, desaturated palette with restrained low-key contrast." ], "visual_inferences": [ "The lighting remains consistently dim, utilizing shadows to emphasize the isolation of the location." ], "creative_decisions": [ "Applying a cool color grade across all shots to unify the visual tone and heighten the desolate, tense mood of the scene." ], "confidence": "high" }, { "element_id": "suri_young_running", "element_type": "character_state", "character_name": "수리영", "description": "A young Korean woman is seen entirely from behind, captured mid-run as she moves rapidly away into the distance. Her back is turned to the camera, and her posture clearly indicates hasty physical movement away from the foreground.", "applies_to_shots": [ 5, 6 ], "element_scope": "full", "source_facts": [ "수리영: back_to_camera, caught mid-run away, gaze=off_screen", "수리영: back_to_camera, running farther away, gaze=off_screen" ], "visual_inferences": [ "Her limbs are in an active running motion, firmly establishing her departure." ], "creative_decisions": [ "Keeping her turned completely away from the lens to emphasize her emotional distance and the suddenness of her leaving." ], "confidence": "high" }, { "element_id": "hye_soo_shouting", "element_type": "character_state", "character_name": "혜수", "description": "An adult Korean woman stands with both hands cupped around her mouth, projecting her voice toward a distant figure. Her upper body is engaged in the act of shouting, and her gaze is fixed forward on the person running away.", "applies_to_shots": [ 5, 6 ], "element_scope": "full", "source_facts": [ "혜수: hands cupped around mouth, gaze=looks_at_character→C01", "혜수: hands held near mouth, gaze=looks_at_character→C01" ], "visual_inferences": [ "Her body weight leans slightly forward in the direction of her voice, physically anchoring her action to the departing figure." ], "creative_decisions": [ "Using her cupped hands as a persistent visual silhouette to instantly communicate her calling out across the distance." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 15, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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": "C02O02", "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" ] }, "render_contracts": [], "_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": "e6603966b99e949a" }, { "scene_index": 15, "heading": "S#15. 해안가. 버스정류장 부근 / N", "beat_title": "관찰자의 시선", "representative_moment": "어두운 구조물 뒤에서 멀리 떨어져 있는 혜수(C02)와 수리영(C01)을 지켜보는 은밀한 시선", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "concealed first-person viewpoint at standing eye level, wide shot with cool desaturated night grading", "t2i_prompt": "Photorealistic cinematic still. A concealing structure edge heavily occludes the middle-left foreground, while C02O02 in a casual leather jacket, a Korean mid-40s woman, and C01O01 in a casual hooded jumper, a Korean 21-year-old woman, appear at a distance in the middle-right background. They both occupy the same distant coastal bus stop vicinity, observed from a static standing eye-level viewpoint. C02O02 stands in right profile, both hands held near her mouth, her eyes fixed on C01O01. Further along the shared path, C01O01 is seen with her back to the camera, captured mid-run as she moves rapidly away into the distance, looking off-screen. Low night ambience and a cool, desaturated color palette keep the near obstruction dense and dark, preserving the sparse, atmospheric outdoor setting.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C02", "outlook_id": "O02" } ], "source_facts": [ "S#15. 해안가. 버스정류장 부근 / N", "수리영: back_to_camera, running farther away, gaze=off_screen, state=alive", "혜수: profile_right, hands held near mouth, gaze=looks_at_character→C01, state=alive" ], "visual_inferences": [ "fixed_element coastal_night_environment: cool, desaturated color palette and restrained low-key lighting", "fixed_element suri_young_running: captured mid-run as she moves rapidly away", "fixed_element hye_soo_shouting: both hands cupped around her mouth", "shared space anchor established between the two distant figures" ], "creative_decisions": [ "Positioned a dark structure edge dominating the left foreground to physically manifest the concealed, subjective POV without showing an observer", "Used a wide framing scale to capture both the departure action and the spatial tension from a distance" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002" ], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "84a1381c2c6cb801", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "640ebf61d8d8904b", "owned_usage_hash": "4f53cda18c2baa0c", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 15, "dependency_reason": "이전 샷(Shot 5)에서 설정된 해안가 버스정류장 주변의 일관된 배경(exact_background) 유지", "_shot_index": 6, "visible_entities": [ "L11", "C01", "C02" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 15, "shot_index": 6 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "wide", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Cut to a static, concealed first-person viewpoint at standing eye level, with one edge of the frame heavily occluded by the dark structure and 혜수 and 수리영 reduced to distant figures. Keep the obstruction close and soft while the two women remain readable across the open space, creating the sense that the camera is deliberately withholding its observer.", "lighting_mood": "Low night ambience and a cool, desaturated grade keep the concealed foreground dense while preserving the distant silhouettes.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "required_background_position", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "foreground", "gesture_action": "none", "gesture_target_label": "", "label": "concealing structure edge", "screen_zone": "middle_left", "target_id": "", "target_kind": "background" }, { "constraint_id": "fsc_002", "depth_plane": "background", "gesture_action": "none", "gesture_target_label": "", "label": "혜수 and 수리영 at a distance", "screen_zone": "middle_right", "target_id": "", "target_kind": "background" } ] }, "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": [ "L11", "C01", "C02" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C02", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "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" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "coastal_night_environment", "element_type": "environment_state", "character_name": "", "description": "The scene is set in a dark coastal area at night, characterized by a cool, desaturated color palette and restrained low-key lighting. The ambient darkness of the seaside dominates the background, establishing a sparse and atmospheric outdoor setting.", "applies_to_shots": [ 1, 5, 6 ], "element_scope": "full", "source_facts": [ "S#15. 해안가. 버스정류장 부근 / N", "Dark coastal night ambience is rendered in a cool, desaturated palette with restrained low-key contrast." ], "visual_inferences": [ "The lighting remains consistently dim, utilizing shadows to emphasize the isolation of the location." ], "creative_decisions": [ "Applying a cool color grade across all shots to unify the visual tone and heighten the desolate, tense mood of the scene." ], "confidence": "high" }, { "element_id": "suri_young_running", "element_type": "character_state", "character_name": "수리영", "description": "A young Korean woman is seen entirely from behind, captured mid-run as she moves rapidly away into the distance. Her back is turned to the camera, and her posture clearly indicates hasty physical movement away from the foreground.", "applies_to_shots": [ 5, 6 ], "element_scope": "full", "source_facts": [ "수리영: back_to_camera, caught mid-run away, gaze=off_screen", "수리영: back_to_camera, running farther away, gaze=off_screen" ], "visual_inferences": [ "Her limbs are in an active running motion, firmly establishing her departure." ], "creative_decisions": [ "Keeping her turned completely away from the lens to emphasize her emotional distance and the suddenness of her leaving." ], "confidence": "high" }, { "element_id": "hye_soo_shouting", "element_type": "character_state", "character_name": "혜수", "description": "An adult Korean woman stands with both hands cupped around her mouth, projecting her voice toward a distant figure. Her upper body is engaged in the act of shouting, and her gaze is fixed forward on the person running away.", "applies_to_shots": [ 5, 6 ], "element_scope": "full", "source_facts": [ "혜수: hands cupped around mouth, gaze=looks_at_character→C01", "혜수: hands held near mouth, gaze=looks_at_character→C01" ], "visual_inferences": [ "Her body weight leans slightly forward in the direction of her voice, physically anchoring her action to the departing figure." ], "creative_decisions": [ "Using her cupped hands as a persistent visual silhouette to instantly communicate her calling out across the distance." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 15, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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": "C02O02", "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" ] }, "render_contracts": [], "_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": "e720e1606e3d2f91" }, { "scene_index": 16, "heading": "S#16. 경찰서 사무실 / N", "beat_title": "불길한 예감에 사로잡힌 혜수", "representative_moment": "혜수(C02)가 사무실로 들어서며 굳은 표정으로 전화를 귀에 대고 있다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Chest-height observer perspective tracking backward, medium framing, cool and desaturated office lighting with restrained contrast", "t2i_prompt": "Photorealistic cinematic still. C02O02 in a leather jacket, a Korean middle-aged woman with a mature oval face and black hair, is positioned in the middle-center midground moving toward the open police office interior. Caught mid-stride planting her leading foot inside, she is viewed from a chest-height angle in three-quarter left profile, entering through the open door behind her. She holds P01, a handheld smartphone, pressed tightly to her ear, the screen facing away from the camera. Her eyes are fixed off-screen with a tense, rigid expression. The stark environment is illuminated by a neutral overhead fluorescent light graded cool and desaturated, casting a sterile mood with restrained contrast across her face, leaving the empty depth of the room extending ahead.", "outfit_assignments": [ { "character_id": "C02", "outlook_id": "O02" } ], "source_facts": [ "S#16. 경찰서 사무실 / N", "사무실 안으로 들어오면서 민숙에게 전화를 거는 혜수", "계속 무응답이다. 왠지 불길한." ], "visual_inferences": [ "fixed_element office_night_lighting: The interior is illuminated by neutral overhead lighting graded cool and desaturated", "Hye-su's body language is rigid and tense caught mid-stride", "The smartphone is pressed to her ear with the screen facing away from the camera" ], "creative_decisions": [ "Anchored the composition with a chest-height camera placing C02 in the middle-center midground to emphasize her entry", "Included the 'door' and 'fluorescent light' from the owned background objects to ground the stark, cool night shift mood" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001" ], "reference_phrase_kinds": [ "character", "background" ], "owned_object_usage": [ { "owned_token": "blind", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "desk", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "desk lamp", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "document tray", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "anchor", "source_phrase": "entering through the open door behind her" }, { "owned_token": "duty desk", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "filing cabinet", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "fluorescent light", "usage_kind": "anchor", "source_phrase": "illuminated by a neutral overhead fluorescent light graded cool and desaturated" }, { "owned_token": "keyboard", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "monitor", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "notice board", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "office chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "partition", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "printer", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "telephone", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wall clock", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wastebasket", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "99695a253a6a0922", "owned_hash": "981aba4a7cdba2cc", "camera_direction_hash": "4f726de64b0e73d3", "owned_usage_hash": "d795eebf7824db19", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "door", "violating_phrase": "entering through the open door behind her", "reason": "The prompt references the door purely as a spatial anchor for the character's entry rather than instructing to redraw or modify the door, which matches the producer's anchor declaration.", "verdict": "anchor_reference" }, { "owned_object": "fluorescent light", "violating_phrase": "illuminated by a neutral overhead fluorescent light graded cool and desaturated", "reason": "The prompt mentions the fluorescent light to describe the ambient lighting and mood rather than to generate a new fixture, aligning with the producer's anchor declaration.", "verdict": "anchor_reference" } ] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "", "_shot_index": 1, "visible_entities": [ "L12", "C02", "P01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 16, "shot_index": 1 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Just inside the office entrance at chest height, track backward as 혜수 plants her leading foot inside, holding briefly on her rigid three-quarter face and the phone pressed to her ear. Frame her in the middle foreground with the entrance behind her and leave open office space ahead for the later curve toward the duty desk.", "lighting_mood": "Neutral office illumination appropriate to the setting is graded cool and desaturated with restrained contrast across 혜수's tense face.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "movement_direction", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "midground", "gesture_action": "moves_toward", "gesture_target_label": "police office interior", "label": "혜수", "screen_zone": "middle_center", "target_id": "C02", "target_kind": "character" } ] }, "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": [ "L12", "C02", "P01" ], "allowed_outlook_pairs": [ { "character_id": "C02", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L12B03", "reference_usage": "exact_background", "owned_objects": [ "blind", "desk", "desk lamp", "document tray", "door", "duty desk", "filing cabinet", "fluorescent light", "keyboard", "monitor", "notice board", "office chair", "partition", "printer", "telephone", "wall clock", "wastebasket", "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "office_night_lighting", "element_type": "environment_state", "character_name": "", "description": "The interior of the police station office is illuminated by neutral overhead lighting suitable for a late-night duty shift. This environmental lighting remains consistently cool and desaturated throughout the area. The overall subdued and stark illumination casts a slightly tense and sterile mood over the space without strong contrast.", "applies_to_shots": [ 1, 4 ], "element_scope": "full", "source_facts": [ "S#16. 경찰서 사무실 / N", "사무실 안에는 김형사 혼자 당직 근무 중이다.", "Neutral office illumination appropriate to the setting is graded cool and desaturated", "Neutral office light is kept cool and desaturated" ], "visual_inferences": [ "The stark, cool tone reflects the late hour of the night shift and heightens the underlying tension of the missing person situation." ], "creative_decisions": [ "Restraining the lighting contrast to keep the focus on the characters' anxious and displeased expressions while maintaining a bleak background." ], "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C02O02", "policy": "required" }, { "kind": "background", "id": "L12B03", "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" ] }, "render_contracts": [], "_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": "bd7f91a2dc1a787a" }, { "scene_index": 16, "heading": "S#16. 경찰서 사무실 / N", "beat_title": "불길한 예감과 위치 추적 지시", "representative_moment": "A close-up of C05 looking down at P01 with a displeased expression.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "tight close view at eye level, cool desaturated lighting", "t2i_prompt": "Photorealistic cinematic still. A tight close view at eye level. An East Asian man in his 30s with a square face and neat black hair, in a worn field jacket, seen at a three-quarter angle from the right, hunched forward. His face dominates the frame, showcasing pursed lips and a displeased expression, gazing off-screen downward. The upper edge of P01 is visible near the lower frame boundary, its illuminated front face angled toward the man. [: The near edge of a duty desk crosses the lower frame beneath the phone, the immediate frame-edge space bathed in cool, desaturated neutral overhead light, with restrained low-key shaping around the man's eyes and mouth.]", "outfit_assignments": [], "source_facts": [ "S#16. 경찰서 사무실 / N", "사무실 안에는 김형사 혼자 당직 근무 중이다.", "shot description: 휴대전화 화면으로 시선을 고정한 채 입술을 삐죽인 상태인 김형사의 못마땅한 표정" ], "visual_inferences": [ "fixed_element office_night_lighting: consistently cool and desaturated throughout the area", "duty desk is in use beneath the phone", "restrained low-key shaping highlights the pursed lips and downward eyeline" ], "creative_decisions": [ "tight close framing to emphasize the subtle displeased expression", "retaining only the upper edge of P01 at the bottom frame to connect his eyeline to the device without breaking the close portrait focus" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [], "reference_phrase_kinds": [], "owned_object_usage": [ { "owned_token": "blind", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "desk", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "desk lamp", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "document tray", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "duty desk", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "filing cabinet", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "fluorescent light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "keyboard", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "monitor", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "notice board", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "office chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "partition", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "printer", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "telephone", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wall clock", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wastebasket", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "6c9f9c1616706df4", "owned_hash": "981aba4a7cdba2cc", "camera_direction_hash": "18fafb78d2347a77", "owned_usage_hash": "2ea45ac7bcfdae18", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 16, "dependency_reason": "Same police station office; requires environmental lighting and mood consistency from the prior shot.", "_shot_index": 4, "visible_entities": [ "C05", "P01" ], "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": "Settle beside the duty desk at 김형사's eye level, angled from the phone side, and hold a tight close view that prioritizes his fixed downward eyeline and pursed lips. Retain the upper edge of the phone screen near the lower frame boundary so his displeasure remains physically connected to what he is reading.", "lighting_mood": "Neutral office light is kept cool and desaturated, with restrained low-key shaping around 김형사's eyes and tightened mouth.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": null, "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", "P01" ], "allowed_outlook_pairs": [ { "character_id": "C05", "outlook_id": "O05" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [ { "subject_id": "C05", "policy_type": "identity_reference", "policy": "generic_descriptor_allowed", "reason": "visible_shot_count=1 — selected-shot 저빈도 주변 subject" } ], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "L12B03", "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", "describe the close subject and its immediate surroundings without relying on the absent background reference", "use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "office_night_lighting", "element_type": "environment_state", "character_name": "", "description": "The interior of the police station office is illuminated by neutral overhead lighting suitable for a late-night duty shift. This environmental lighting remains consistently cool and desaturated throughout the area. The overall subdued and stark illumination casts a slightly tense and sterile mood over the space without strong contrast.", "applies_to_shots": [ 1, 4 ], "element_scope": "full", "source_facts": [ "S#16. 경찰서 사무실 / N", "사무실 안에는 김형사 혼자 당직 근무 중이다.", "Neutral office illumination appropriate to the setting is graded cool and desaturated", "Neutral office light is kept cool and desaturated" ], "visual_inferences": [ "The stark, cool tone reflects the late hour of the night shift and heightens the underlying tension of the missing person situation." ], "creative_decisions": [ "Restraining the lighting contrast to keep the focus on the characters' anxious and displeased expressions while maintaining a bleak background." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 16, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [], "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" ] }, "render_contracts": [], "_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": "1270cc4164970cdc" }, { "scene_index": 17, "heading": "S#17. 옥탑방 / 동틀무렵", "beat_title": "새벽녘 옥탑방을 주시하는 검은 고양이", "representative_moment": "동이 트는 푸른 새벽, 검은 고양이가 불 켜진 옥탑방 창문을 응시하고 있다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "wide composition from cat's eye level, pale blue dawn ambience", "t2i_prompt": "Photorealistic cinematic still. From the eye level of the existing black cat in the lower-left foreground, looking slightly from behind its shoulder toward the lit rooftop-room window in the upper-right background. An unbroken line of sight aligns the cat's gaze with the window across the shared rooftop. [L03: an old multi-family villa rooftop enveloped in pale blue dawn ambience, the lit rooftop-room window facing the camera obliquely as a restrained contrasting accent, with lateral space opening toward a nearby parapet.]", "outfit_assignments": [], "source_facts": [ "S#17 action: 'A black cat looks at the lit rooftop room'", "shot description: black cat fixing its gaze on the lit rooftop-room window under pale blue dawn ambience" ], "visual_inferences": [ "static wide composition from slightly behind and to one side", "three-quarter left angle, forepaws planted, neck craned, gaze off-screen toward the window", "unbroken line of sight between the cat and the window" ], "creative_decisions": [ "placing the feline low in the lower-left foreground to establish depth", "pale blue dawn shadows contrasting with the warm restrained accent of the lit window" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002" ], "reference_phrase_kinds": [ "background" ], "owned_object_usage": [ { "owned_token": "AC", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "building", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "cat", "usage_kind": "anchor", "source_phrase": "the existing black cat in the lower-left foreground" }, { "owned_token": "clothesline", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "flowerpot", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "gas meter", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "laundry", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "parapet", "usage_kind": "anchor", "source_phrase": "nearby parapet" }, { "owned_token": "pipe", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "rooftop", "usage_kind": "anchor", "source_phrase": "across the shared rooftop" }, { "owned_token": "rooftop room", "usage_kind": "anchor", "source_phrase": "rooftop-room window" }, { "owned_token": "utility pole", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "water tank", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "anchor", "source_phrase": "lit rooftop-room window in the upper-right background" }, { "owned_token": "wire", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "5a7f55c2baf7766c", "owned_hash": "9b79eace1cc73e5d", "camera_direction_hash": "6c4ea5a053f103eb", "owned_usage_hash": "8e2b1ee84d9cfd45", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "cat", "violating_phrase": "the existing black cat in the lower-left foreground", "reason": "The prompt references the existing black cat only to establish the camera's eye level and composition, matching the producer's anchor classification.", "verdict": "anchor_reference" }, { "owned_object": "parapet", "violating_phrase": "nearby parapet", "reason": "The prompt uses the parapet purely as a spatial composition anchor to define lateral space, matching the producer's anchor classification.", "verdict": "anchor_reference" }, { "owned_object": "rooftop", "violating_phrase": "across the shared rooftop", "reason": "The prompt mentions the rooftop as a spatial reference for the line of sight and scene setting, matching the producer's anchor classification.", "verdict": "anchor_reference" }, { "owned_object": "rooftop room", "violating_phrase": "rooftop-room window", "reason": "The prompt mentions the rooftop-room as a modifier for the window's location, matching the producer's anchor classification.", "verdict": "anchor_reference" }, { "owned_object": "window", "violating_phrase": "lit rooftop-room window", "reason": "The prompt describes the window as lit, which is a visual state change explicitly required by the shot_intent ('불 켜진 옥탑방 창문'), satisfying the exception for allowed visual state change.", "verdict": "allowed_visual_state_change" } ] } } ], "scene_type": "normal", "dependent_scene_index": 10, "dependency_reason": "연속성 유지: 이전 씬(S10_Shot3)과 동일한 옥탑방 배경을 참조하여 정확한 위치감을 형성", "_shot_index": 1, "visible_entities": [ "L03", "C12" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 17, "shot_index": 1 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "wide", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From the black cat's eye level, hold a static wide composition from slightly behind and to one side, placing the cat low in the foreground and aiming past its head toward the lit rooftop-room window. Preserve the unbroken line between its rigid gaze and the window while leaving lateral space in the direction of the wall it will approach next.", "lighting_mood": "Pale blue dawn ambience surrounds the black cat while the lit window remains a restrained contrasting accent.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "looks_to_anchor", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "background", "gesture_action": "none", "gesture_target_label": "", "label": "lit rooftop-room window", "screen_zone": "upper_right", "target_id": "", "target_kind": "background" }, { "constraint_id": "fsc_002", "depth_plane": "foreground", "gesture_action": "looks_toward", "gesture_target_label": "lit rooftop-room window", "label": "검은 고양이", "screen_zone": "lower_left", "target_id": "C12", "target_kind": "character" } ] }, "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": [ "L03", "C12" ], "allowed_outlook_pairs": [ { "character_id": "C12", "outlook_id": "O00" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [ { "subject_id": "C12", "policy_type": "identity_reference", "policy": "generic_descriptor_allowed", "reason": "visible_shot_count=1 — selected-shot 저빈도 주변 subject" } ], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L03B01", "reference_usage": "exact_background", "owned_objects": [ "AC", "building", "cat", "clothesline", "door", "flowerpot", "gas meter", "laundry", "parapet", "pipe", "rooftop", "rooftop room", "utility pole", "water tank", "window", "wire" ], "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [ { "scene_index": 10, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "background", "id": "L03B01", "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" ] }, "render_contracts": [], "_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": "b4f2e28d42186bf4" }, { "scene_index": 18, "heading": "S#18. 안방 안 / D", "beat_title": "고양이 울음소리에 깨어나 불안한 직감에 휩싸인 수리영", "representative_moment": "침대에 기대어 잠들었던 C01이 고양이 소리에 번쩍 눈을 뜨고 굳은 표정으로 허공을 응시한다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "seated eye level close framing with cool desaturated palette", "t2i_prompt": "Photorealistic cinematic still. C01O01 in a casual hooded jumper, a 21-year-old Asian woman, isolated in an awakening close view in the middle-center foreground. Captured at an observer's seated eye level beside the bed, her face occupies most of the frame in a tight three-quarter angle from the left. Her head is just lifted from the mattress, eyes snapped open in a rigid stare, looking off-screen toward an unseen concern. Only a small portion of the bed's near upper edge remains visible at the lower frame boundary, anchoring her seated posture. The ambient indoor illumination features a cool, desaturated color palette and restrained low-key contrast, casting a bleak and unsettling atmosphere across her tense facial features.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "S#18 action: '침대에 머리를 대고 앉아있는 수리영, 깜박 졸았다 고양이 울음소리에 번쩍 눈을 뜨고'", "shot description: observer at seated eye level beside the bed, tight three-quarter view, face occupies most of the frame, gazing off-screen." ], "visual_inferences": [ "fixed_element cool_desaturated_lighting: The lighting retains a cool, desaturated palette with low-key contrast to build dread.", "A small slice of the bed anchors the bottom edge of the frame to show where she fell asleep, while the background remains absent due to close framing." ], "creative_decisions": [ "Eliminated background room elements entirely to strictly follow the close framing policy, focusing the tension entirely on her facial expression and wide eyes." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001" ], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [ { "owned_token": "bathroom door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "bathroom mirror", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "dining table", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "doorway", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "photograph", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "red circle", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "d1cd46015b47c1a4", "owned_hash": "e0e0bf1ca9e1aaf8", "camera_direction_hash": "8c63fd9effecf1d5", "owned_usage_hash": "797b8500e9e1b8fa", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 14, "dependency_reason": "이전 샷에서 설정된 차갑고 채도가 낮은 실내 조명(cool desaturated lighting)의 톤앤매너 연속성 유지", "_shot_index": 2, "visible_entities": [ "C01" ], "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": "At seated eye level beside the bed, finish the dolly-in on a tight three-quarter view of 수리영 as her eyes snap open and lock toward the unseen concern. Let her face occupy most of the frame while a small portion of the bed remains at the lower edge, holding the rigid stare before the camera eases back for her rise.", "lighting_mood": "Natural ambient daytime light appropriate to the room is treated with a cool, desaturated palette and restrained contrast around her suddenly opened eyes.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "primary_subject_isolation", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "foreground", "gesture_action": "none", "gesture_target_label": "", "label": "수리영 isolated in awakening close view", "screen_zone": "middle_center", "target_id": "C01", "target_kind": "character" } ] }, "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" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "L05B01", "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", "describe the close subject and its immediate surroundings without relying on the absent background reference", "use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "cool_desaturated_lighting", "element_type": "environment_state", "character_name": "", "description": "The ambient indoor lighting remains consistent across the different rooms, featuring a cool and desaturated color palette. The illumination is characterized by restrained, low-key contrast that casts a bleak and unsettling atmosphere without relying on harsh directional light.", "applies_to_shots": [ 2, 8, 12 ], "element_scope": "full", "source_facts": [ "Natural ambient daytime light appropriate to the room is treated with a cool, desaturated palette and restrained contrast around her suddenly opened eyes (Shot 2).", "Ambient bathroom light is kept cool and desaturated (Shot 8).", "Use restrained low-key contrast and a cool desaturated grade, without introducing a specific unestablished light source (Shot 12)." ], "visual_inferences": [ "The unified color grading across the bedroom, bathroom, and living room establishes a continuous sense of psychological dread and isolation." ], "creative_decisions": [ "Maintained the same desaturated, low-key tone across disparate physical locations within the house to emphasize the surreal nature of the hallucinations." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 14, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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" } ], "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" ] }, "render_contracts": [], "_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": "a5781ac10b6add0e" }, { "scene_index": 18, "heading": "S#18. 안방 안 및 욕실 안 / D", "beat_title": "거울 속 환각과 자기통제", "representative_moment": "욕실에서 세수를 하던 수리영(C01)이 등 뒤 거울에 나타난 기괴한 붉은 원을 목격한다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Dolly in from directly behind at shoulder height, medium framing, exact background reflection", "t2i_prompt": "Photorealistic cinematic still. C01O01 in a casual hooded jacket, a 21-year-old Korean woman, stands in the lower-left foreground with her back to the camera, facing the bathroom mirror which shows the reflected bathroom wall with red circle in the middle-right background. Captured from directly behind at shoulder height, at least one vertical edge of the bathroom mirror is visible on the left to establish the reversed view. She leans forward toward the glass. Inside the reflective surface, the reflected face and torso of the young woman occupy the middle-left portion, her eyes looking off-screen. The reflected tile wall extends broadly to the right, cleanly separating her reflection from the deep-crimson mark. The ambient indoor lighting is kept cool and desaturated with restrained low-key contrast, allowing the vivid red mark to register as a single forceful color accent in the bleak room.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "Suyeong stands in the bathroom with her back to the camera, leaning toward the mirror.", "A sudden red circle appears on the reflected bathroom wall behind her.", "The shot description specifies her physical back is low and left, her reflected figure is middle-left, and the red mark is on the reflected wall with more space on the right." ], "visual_inferences": [ "fixed_element cool_desaturated_lighting: The ambient indoor lighting remains consistent, featuring a cool and desaturated color palette with restrained low-key contrast.", "The mirror acts as a reproduction surface, providing a left-right reversed view of her face and the tile wall.", "The red circle's appearance is isolated to the reflection, emphasizing the supernatural horror aspect." ], "creative_decisions": [ "Framed from directly behind at shoulder height to include both her physical back in the foreground and her reflection, maximizing the eerie separation between reality and hallucination.", "Used deep-crimson for the red mark to create a sharp, unsettling focal point against the cool, desaturated tiles." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002" ], "reference_phrase_kinds": [ "character", "background" ], "owned_object_usage": [ { "owned_token": "bathroom door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "bathroom mirror", "usage_kind": "anchor", "source_phrase": "facing the bathroom mirror" }, { "owned_token": "chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "dining table", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "doorway", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "photograph", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "red circle", "usage_kind": "anchor", "source_phrase": "reflected bathroom wall with red circle" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "80318e4dd0610719", "owned_hash": "e0e0bf1ca9e1aaf8", "camera_direction_hash": "5a9085f2e020216f", "owned_usage_hash": "244995a6e1a4bf60", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "bathroom mirror", "violating_phrase": "facing the bathroom mirror", "reason": "The prompt references the bathroom mirror as a spatial anchor and framing edge to establish the reflection, which aligns with the producer's anchor declaration.", "verdict": "anchor_reference" }, { "owned_object": "red circle", "violating_phrase": "reflected bathroom wall with red circle", "reason": "The prompt references the red circle (and the deep-crimson mark) as a pre-existing visual accent in the reflection, which aligns with the producer's anchor declaration.", "verdict": "anchor_reference" } ] } } ], "scene_type": "normal", "dependent_scene_index": 12, "dependency_reason": "이전 씬의 집안 배경 및 서늘하고 채도 낮은 조명 톤(cool_desaturated_lighting) 연속성 유지", "_shot_index": 8, "visible_entities": [ "L05", "C01", "P02" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 18, "shot_index": 8 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Dolly in from directly behind 수리영 at shoulder height toward the bathroom mirror, keeping her physical back low and left while her reflected figure occupies the middle-left portion of the mirror. Give the reflected wall more space on the right so the vivid red circle remains cleanly separated from her reflection, and retain at least one mirror edge to establish the left-right-reversed reflected view.", "lighting_mood": "Ambient bathroom light is kept cool and desaturated, allowing the reflected deep-crimson circle to register as the single forceful color accent.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "required_background_position", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "background", "gesture_action": "none", "gesture_target_label": "", "label": "reflected bathroom wall with red circle", "screen_zone": "middle_right", "target_id": "", "target_kind": "background" }, { "constraint_id": "fsc_002", "depth_plane": "foreground", "gesture_action": "none", "gesture_target_label": "", "label": "수리영", "screen_zone": "lower_left", "target_id": "C01", "target_kind": "character" } ] }, "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": [ "L05", "C01", "P02" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L05B01", "reference_usage": "exact_background", "owned_objects": [ "bathroom door", "bathroom mirror", "chair", "dining table", "doorway", "photograph", "red circle", "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "cool_desaturated_lighting", "element_type": "environment_state", "character_name": "", "description": "The ambient indoor lighting remains consistent across the different rooms, featuring a cool and desaturated color palette. The illumination is characterized by restrained, low-key contrast that casts a bleak and unsettling atmosphere without relying on harsh directional light.", "applies_to_shots": [ 2, 8, 12 ], "element_scope": "full", "source_facts": [ "Natural ambient daytime light appropriate to the room is treated with a cool, desaturated palette and restrained contrast around her suddenly opened eyes (Shot 2).", "Ambient bathroom light is kept cool and desaturated (Shot 8).", "Use restrained low-key contrast and a cool desaturated grade, without introducing a specific unestablished light source (Shot 12)." ], "visual_inferences": [ "The unified color grading across the bedroom, bathroom, and living room establishes a continuous sense of psychological dread and isolation." ], "creative_decisions": [ "Maintained the same desaturated, low-key tone across disparate physical locations within the house to emphasize the surreal nature of the hallucinations." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 12, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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": "background", "id": "L05B01", "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" ] }, "render_contracts": [], "_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": "330bf712434a7a96" }, { "scene_index": 18, "heading": "S#18. 안방 안 / D", "beat_title": "기괴하게 일그러지는 사진 속 엄마의 얼굴 환각", "representative_moment": "The smiling face of the mother in the photograph P04 distorts and smears downward.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "close framing, slightly downward over-the-shoulder angle, cool desaturated grade", "t2i_prompt": "Photorealistic cinematic still. A close framing looking slightly downward at P04, a framed photograph. The printed front of P04 faces the camera squarely, with only a narrow dark border of the physical frame remaining visible along the outer edges. Inside the photograph, a printed portrait of an Asian middle-aged woman is visible, facing the camera with a smiling expression, but her printed facial features are severely distorted, frozen in a hallucinated downward smear across the flat glossy paper surface. The illumination features restrained low-key contrast and a cool desaturated color palette, catching a faint unsettling sheen on the distorted image. [L05B01: dark wooden table surface softly blurred in the immediate background just beyond the photo edge].", "outfit_assignments": [], "source_facts": [ "S#18 action: '식탁에 놓인 엄마와 수리영의 사진을 보는데... 환하게 웃는 엄마 얼굴이 뭉개져 흘러내린다'", "shot description: Kang Minsuk's smiling face in the framed photo collapses and smears downward." ], "visual_inferences": [ "fixed_element cool_desaturated_lighting: The ambient indoor lighting remains consistent with a cool and desaturated color palette and restrained low-key contrast.", "The camera angle looks slightly downward over the shoulder to fill the frame with the photograph surface." ], "creative_decisions": [ "Used a close framing to emphasize the hallucinated smearing effect on the flat paper surface while keeping the physical frame border visible for context.", "Applied a generic demographic descriptor (an Asian middle-aged woman) for the reproduced face in the photograph as per the reproduction_surface_rule." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [], "reference_phrase_kinds": [ "prop" ], "owned_object_usage": [ { "owned_token": "bathroom door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "bathroom mirror", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "dining table", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "doorway", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "photograph", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "red circle", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "ce12c8803130df87", "owned_hash": "e0e0bf1ca9e1aaf8", "camera_direction_hash": "f81fdb9a6c4e1e6b", "owned_usage_hash": "797b8500e9e1b8fa", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 12, "dependency_reason": "Maintained lighting consistency (cool desaturated low-key) from prior hallucination/dread moments established in the house.", "_shot_index": 12, "visible_entities": [ "P04" ], "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": "End the inherited dolly-in directly over the framed photograph, slightly above and behind where 수리영's shoulder had been, until 강민숙's smiling face occupies most of the close frame while a narrow strip of the photograph edge remains visible. Hold at the peak of the hallucinated facial collapse, rendering the printed features as if they are smearing downward within the image rather than transforming into a physically present person.", "lighting_mood": "Use restrained low-key contrast and a cool desaturated grade, without introducing a specific unestablished light source.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": null, "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" ], "allowed_outlook_pairs": [], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "L05B01", "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", "describe the close subject and its immediate surroundings without relying on the absent background reference", "use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "cool_desaturated_lighting", "element_type": "environment_state", "character_name": "", "description": "The ambient indoor lighting remains consistent across the different rooms, featuring a cool and desaturated color palette. The illumination is characterized by restrained, low-key contrast that casts a bleak and unsettling atmosphere without relying on harsh directional light.", "applies_to_shots": [ 2, 8, 12 ], "element_scope": "full", "source_facts": [ "Natural ambient daytime light appropriate to the room is treated with a cool, desaturated palette and restrained contrast around her suddenly opened eyes (Shot 2).", "Ambient bathroom light is kept cool and desaturated (Shot 8).", "Use restrained low-key contrast and a cool desaturated grade, without introducing a specific unestablished light source (Shot 12)." ], "visual_inferences": [ "The unified color grading across the bedroom, bathroom, and living room establishes a continuous sense of psychological dread and isolation." ], "creative_decisions": [ "Maintained the same desaturated, low-key tone across disparate physical locations within the house to emphasize the surreal nature of the hallucinations." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 12, "shot_index": 8, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "prop", "id": "P04", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "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" ] }, "render_contracts": [ { "contract_id": "rc_001", "scope": { "scene_index": 18, "shot_indices": [ 12 ], "duration": "single_shot" }, "targets": [ { "entity_id": "P04", "role": "visual_target" } ], "requirements": [ { "dimension": "visual_identity", "operation": "preserve", "strength": "required", "reference_policy": "use_entity_reference" } ] } ], "_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": "26dbf2c5b2204ded" }, { "scene_index": 19, "heading": "S#19. 마트 / D", "beat_title": "Searching for clues in the mart", "representative_moment": "C01 is caught mid-stride in the mart aisle, searching urgently for the male employee.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Wide chest-height tracking shot with cool, desaturated early-morning ambient lighting", "t2i_prompt": "Photorealistic cinematic still. C01O01 in a casual hooded jacket, a Korean young woman with tied-up black hair, is caught mid-stride in the middle-center midground, moving toward the far end of the store aisle in the upper-center background. Captured from chest height in wide framing, her full body is visible. She appears at a three-quarter angle from the right, her head turned in an urgent search, eyes fixed off-screen. Parallel shelves flank the aisle on both sides, receding into the distance over the tile floor. The scene is lit by cool, desaturated ambient store illumination from ceiling lights, creating a restrained early-morning mood.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "S#19 action: '마트 여기저기를 돌아다니며 어제 만났던 남자 직원을 찾는다'", "shot description: C01 looking around carefully in the aisle of a quiet mart in the early morning" ], "visual_inferences": [ "wide framing captures full body context to emphasize the quiet scale of the mart", "chest-height angle anchors the subject naturally within the aisle layout" ], "creative_decisions": [ "cool, desaturated ambient light emphasizes the early-morning desolation of the mart", "parallel shelves and tile floor provide leading lines to structure the tracking perspective" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002" ], "reference_phrase_kinds": [ "character", "background" ], "owned_object_usage": [ { "owned_token": "ceiling light", "usage_kind": "anchor", "source_phrase": "from ceiling lights" }, { "owned_token": "price tag", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "produce island", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "product", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "refrigerated display case", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "security camera", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "shelf", "usage_kind": "anchor", "source_phrase": "Parallel shelves flank the aisle" }, { "owned_token": "shopping cart", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "sign", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "tile floor", "usage_kind": "anchor", "source_phrase": "over the tile floor" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "31a85f2b3e62db33", "owned_hash": "0a5aa7dc6a17202f", "camera_direction_hash": "6ec07414400c2913", "owned_usage_hash": "f5598b3d7a9050d0", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "ceiling light", "violating_phrase": "from ceiling lights", "reason": "The prompt references the ceiling lights purely as the source of ambient store illumination, aligning with the producer's anchor declaration.", "verdict": "anchor_reference" }, { "owned_object": "shelf", "violating_phrase": "Parallel shelves flank the aisle", "reason": "The prompt references parallel shelves to establish the layout and composition of the background, matching the producer's anchor declaration.", "verdict": "anchor_reference" }, { "owned_object": "tile floor", "violating_phrase": "over the tile floor", "reason": "The prompt mentions the tile floor to describe the perspective of the receding aisle, serving as a spatial anchor in agreement with the producer's declaration.", "verdict": "anchor_reference" } ] } } ], "scene_type": "normal", "dependent_scene_index": 8, "dependency_reason": "Same mart interior background (L09), maintaining exact background consistency from previous visit.", "_shot_index": 1, "visible_entities": [ "L09", "C01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 19, "shot_index": 1 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "wide", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From chest height five meters ahead in the aisle, track backward at 수리영's pace while maintaining a full-body view between the displays. Keep her slightly off-center with the aisle extending behind her, catching her mid-stride as her head and eyes search from one side to the other.", "lighting_mood": "Use ambient store illumination appropriate to the early-morning interior, graded cool and desaturated with restrained contrast.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "movement_direction", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "background", "gesture_action": "none", "gesture_target_label": "", "label": "far end of the store aisle", "screen_zone": "upper_center", "target_id": "", "target_kind": "background" }, { "constraint_id": "fsc_002", "depth_plane": "midground", "gesture_action": "moves_toward", "gesture_target_label": "far end of the store aisle", "label": "수리영", "screen_zone": "middle_center", "target_id": "C01", "target_kind": "character" } ] }, "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", "C01" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L09B04", "reference_usage": "exact_background", "owned_objects": [ "ceiling light", "price tag", "produce island", "product", "refrigerated display case", "security camera", "shelf", "shopping cart", "sign", "tile floor" ], "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [ { "scene_index": 8, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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": "background", "id": "L09B04", "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" ] }, "render_contracts": [], "_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": "37b6ea529827221e" }, { "scene_index": 19, "heading": "S#19. 마트 / D", "beat_title": "CCTV 위치를 발견하는 수리영", "representative_moment": "마트 건물 외벽의 CCTV를 다급하게 올려다보는 수리영의 얼굴", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "restrained low angle with shallow focus", "t2i_prompt": "Photorealistic cinematic still. C01O01, an Asian woman in her 20s in a casual hood jumper, stands in the middle-left foreground, looking up toward an exterior CCTV camera in the upper-right background. Shot from an observer perspective slightly below her eye level, her upturned face occupies most of the lower and middle frame at a three-quarter angle from the left, neck craned upward. The exterior CCTV camera remains clearly visible as a silhouette above her along the same sightline, anchored against [L10: an out-of-focus exterior wall surface]. The scene is captured with shallow focus, illuminated by natural ambient light of an early-morning exterior with a cool desaturated grade and controlled facial contrast.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "S#19 action: '엄마와 남자 직원CCTV 위치를 확인하다. 찍혔을 만한 위치다.'", "shot description: 마트 건물 밖, 고개를 뒤로 젖혀 건물 외벽의 CCTV 카메라를 올려다보는 수리영의 얼굴" ], "visual_inferences": [ "neck craned upward pointing to the high placement of the camera on the wall", "early morning ambient light touching her upturned features" ], "creative_decisions": [ "shallow depth of field isolates the subject and the CCTV from the skipped background", "restrained low angle tightens within a close space, enhancing the suspense of the discovery", "cool desaturated grade supports the grounded, suspenseful morning atmosphere" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002" ], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [ { "owned_token": "AC unit", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "awning", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "cable conduit", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "gas meter", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "laundry line", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "mounting bracket", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "planter", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "power line", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "rolling shutter", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "security bar", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "security camera", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "signboard", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "utility pole", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "vehicle", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "water tank", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "c8476c673bd164d8", "owned_hash": "5b817f8e39139a99", "camera_direction_hash": "0b350e952dc448bd", "owned_usage_hash": "2622fdae9355a99e", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 8, "dependency_reason": "같은 마을 내 마트 외벽의 시각적 톤과 질감을 일관되게 유지하기 위함 (배경이 스킵되더라도 조명과 벽면 표면의 톤 유지)", "_shot_index": 4, "visible_entities": [ "L10", "C01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 19, "shot_index": 4 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Continue the reveal tilt from slightly below 수리영's eye level, tightening within one meter so her upturned face occupies most of the lower and middle frame while the CCTV camera remains above along the same sightline. Use a restrained low angle and shallow focus that preserves the CCTV silhouette clearly enough to establish what she has found.", "lighting_mood": "Use natural ambient light appropriate to the early-morning exterior with a cool desaturated grade and controlled facial contrast.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "looks_to_anchor", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "background", "gesture_action": "none", "gesture_target_label": "", "label": "exterior CCTV camera", "screen_zone": "upper_right", "target_id": "", "target_kind": "background" }, { "constraint_id": "fsc_002", "depth_plane": "foreground", "gesture_action": "looks_toward", "gesture_target_label": "exterior CCTV camera", "label": "수리영", "screen_zone": "middle_left", "target_id": "C01", "target_kind": "character" } ] }, "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": [ "L10", "C01" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "L09B03", "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", "describe the close subject and its immediate surroundings without relying on the absent background reference", "use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [ { "scene_index": 8, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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" } ], "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" ] }, "render_contracts": [], "_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": "313dc6b9d5878dac" }, { "scene_index": 20, "heading": "S#20. 마트 총무과 / 실내", "beat_title": "식별 불가능한 CCTV 화면과 수리영의 불안", "representative_moment": "C02O02 leans toward the dark CCTV monitor while C01O01 watches anxiously from behind in the dim office.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Medium shot pushing along the monitor axis, observer POV with cool-desaturated low-key lighting", "t2i_prompt": "Photorealistic cinematic still. C02O02 in a leather jacket, an Asian middle-aged woman, leans forward in the middle-left foreground looking toward the CCTV monitor in the upper-center midground, while C01O01 in a casual hoodie, an Asian 21-year-old woman with long black hair, hovers in the middle-right background looking toward the CCTV monitor, both sharing the dim workspace across the same desk. Captured in an over-the-shoulder perspective with her face partially visible in profile, C02O02 anchors the foreground, while C01O01 stands behind an operator chair at a three-quarter left angle. The active front display of the CCTV monitor faces obliquely toward the camera, showing a murky, unusually dark surveillance image. Subdued, cool-desaturated ambient light washes over the office, maintaining low-key contrast and preserving the somber atmosphere without any conspicuous extra light sources.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C02", "outlook_id": "O02" } ], "source_facts": [ "The scene is inside the mart general affairs office.", "C02 (혜수) leans close to the monitor, staring at the screen.", "C01 (수리영) stands behind, hovering anxiously.", "The CCTV monitor displays unusually dark and indiscernible footage." ], "visual_inferences": [ "fixed_element cedar_harbor: office context with monitor-viewing workspace maintained", "fixed_element moss_compass: subdued, cool-desaturated office lighting with low-key contrast", "Shared desk anchors the separated foreground and background figures in the same continuous physical space." ], "creative_decisions": [ "Over-the-shoulder framing for C02 applied to emphasize the shared line of sight toward the monitor screen.", "Applied medium framing scale to comfortably capture both figures and the active screen within the same frame." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002", "fsc_003" ], "reference_phrase_kinds": [ "character", "background" ], "owned_object_usage": [ { "owned_token": "CCTV monitor", "usage_kind": "anchor", "source_phrase": "looking toward the CCTV monitor in the upper-center midground" }, { "owned_token": "blind", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "bulletin board", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "computer monitor", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "desk", "usage_kind": "anchor", "source_phrase": "sharing the dim workspace across the same desk" }, { "owned_token": "desktop computer", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "filing cabinet", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "fluorescent light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "keyboard", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "mouse", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "operator chair", "usage_kind": "anchor", "source_phrase": "stands behind an operator chair at a three-quarter left angle" }, { "owned_token": "printer", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "shelf", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "telephone", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "visitor chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wall clock", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "627903eb7daeac41", "owned_hash": "950c0cdce47336fd", "camera_direction_hash": "34e5cc51e86eed75", "owned_usage_hash": "ac417d77294c0f75", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "CCTV monitor", "violating_phrase": "The active front display of the CCTV monitor faces obliquely toward the camera, showing a murky, unusually dark surveillance image.", "reason": "The prompt specifies the depicted image on the screen, which is permitted as a visual state change since the shot_intent structurally requires a 'dark CCTV monitor'.", "verdict": "allowed_visual_state_change" }, { "owned_object": "desk", "violating_phrase": "sharing the dim workspace across the same desk", "reason": "The desk is referenced solely as a spatial anchor for the characters, matching the producer's anchor declaration.", "verdict": "anchor_reference" }, { "owned_object": "operator chair", "violating_phrase": "stands behind an operator chair at a three-quarter left angle", "reason": "The operator chair is mentioned only to position the character relative to it, matching the producer's anchor declaration.", "verdict": "anchor_reference" } ] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "", "_shot_index": 1, "visible_entities": [ "L13", "C01", "C02" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 20, "shot_index": 1 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Begin at chest height behind and slightly beside 혜수, then push forward along the shared monitor axis while retaining 수리영 farther back in the same frame. 혜수 leans into the left foreground and partially frames the dark CCTV display, while 수리영 remains fully separated in the rear, tense and fixed on the screen.", "lighting_mood": "Keep the office subdued and cool-desaturated, allowing the active monitor to register without inventing additional light sources.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "shared_space_relation", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "midground", "gesture_action": "none", "gesture_target_label": "", "label": "CCTV monitor", "screen_zone": "upper_center", "target_id": "", "target_kind": "background" }, { "constraint_id": "fsc_002", "depth_plane": "background", "gesture_action": "looks_toward", "gesture_target_label": "CCTV monitor", "label": "수리영", "screen_zone": "middle_right", "target_id": "C01", "target_kind": "character" }, { "constraint_id": "fsc_003", "depth_plane": "foreground", "gesture_action": "looks_toward", "gesture_target_label": "CCTV monitor", "label": "혜수", "screen_zone": "middle_left", "target_id": "C02", "target_kind": "character" } ] }, "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", "C01", "C02" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C02", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L13B01", "reference_usage": "exact_background", "owned_objects": [ "CCTV monitor", "blind", "bulletin board", "computer monitor", "desk", "desktop computer", "door", "filing cabinet", "fluorescent light", "keyboard", "mouse", "operator chair", "printer", "shelf", "telephone", "visitor chair", "wall clock" ], "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "applies_to_shots": [ 1, 4 ], "character_name": "", "confidence": "high", "creative_decisions": [], "description": "The scene remains inside the mart general affairs office, presented as a CCTV monitoring workspace. Maintain recognizable office context around the monitor-viewing and employee conversation areas without changing the established location.", "element_id": "cedar_harbor", "element_scope": "full", "element_type": "environment_state", "source_facts": [ "The scene heading identifies the location as the interior of the mart general affairs office.", "Shot 1 places the characters inside the mart general affairs office and CCTV control space.", "Shot 4 retains enough office context to preserve the conversational axis." ], "visual_inferences": [] }, { "applies_to_shots": [ 1, 4 ], "character_name": "", "confidence": "high", "creative_decisions": [], "description": "The office remains dimly and evenly restrained, with a cool, desaturated palette and low-key contrast. Do not introduce conspicuous additional light sources or shift the room toward warm, bright illumination.", "element_id": "moss_compass", "element_scope": "full", "element_type": "environment_state", "source_facts": [ "Shot 1 specifies subdued, cool-desaturated office lighting and instructs against inventing additional light sources.", "Shot 4 specifies restrained office illumination with a cool desaturated palette and low-key contrast." ], "visual_inferences": [] } ], "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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": "C02O02", "policy": "required" }, { "kind": "background", "id": "L13B01", "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" ] }, "render_contracts": [], "_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": "c7b976cbbf351a5a" }, { "scene_index": 20, "heading": "S#20. 마트 총무과 / 실내", "beat_title": "CCTV 원본 영상 복사 요청", "representative_moment": "혜수가 보안팀 직원을 향해 단호한 표정으로 CCTV 원본 복사를 요청하는 순간.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Observer POV close framing, eye-level lateral track, cool desaturated palette", "t2i_prompt": "Photorealistic cinematic still. C02O02 in a casual leather jacket, an Asian middle-aged woman with a mature oval face and black hair, placed in the left-center frame in right profile, turning mid-request with mouth half-open, eyes looking off-screen. A Korean male security employee in worker uniform, seated in the soft background near the right edge at three-quarter angle from the left, listening closely, eyes fixed on C02O02. The side edge of a shared work desk is visible between them in the lower frame, anchoring their space. A CCTV monitor display is seen obliquely from the side in the background, its dark footage faintly visible but illegible. Lit by restrained cool desaturated office illumination with low-key contrast focused on C02O02's resolute profile, against [: dimly lit office walls forming the out-of-focus background edge].", "outfit_assignments": [ { "character_id": "C02", "outlook_id": "O02" } ], "source_facts": [ "The scene heading identifies the location as the interior of the mart general affairs office.", "Shot 4 retains enough office context to preserve the conversational axis.", "Hyesu makes a firm request to copy the entire CCTV footage." ], "visual_inferences": [ "fixed_element cedar_harbor: The scene remains inside the mart general affairs office, presented as a CCTV monitoring workspace.", "fixed_element moss_compass: The office remains dimly and evenly restrained, with a cool, desaturated palette and low-key contrast.", "The CCTV monitor displays dark, illegible footage as evidence of the discussion." ], "creative_decisions": [ "Employed close framing to strictly focus on Hyesu's resolute right profile and mid-request expression.", "Placed the security employee in the soft background to adhere to the close framing primary subject rule, maintaining the conversational axis without competing with the primary focus.", "Used a shared work desk in the lower frame as a spatial anchor to clearly define the foreground-background separation." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [ { "owned_token": "CCTV monitor", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "blind", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "bulletin board", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "computer monitor", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "desk", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "desktop computer", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "filing cabinet", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "fluorescent light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "keyboard", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "mouse", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "operator chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "printer", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "shelf", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "telephone", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "visitor chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wall clock", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "78f011c2e4dac957", "owned_hash": "950c0cdce47336fd", "camera_direction_hash": "25c6473fcc92b629", "owned_usage_hash": "0f7a9bd89a4d056f", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 20, "dependency_reason": "Shot 1에 설정된 쿨하고 채도 낮은 총무과 사무실 조명과 공간적 연속성(Conversational axis) 유지", "_shot_index": 4, "visible_entities": [ "C02" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 20, "shot_index": 4 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Complete the lateral track at 혜수's eye height, settling close across her firm side profile as she turns toward the security employee. Place 혜수 in sharp left-center profile with her half-open mouth caught mid-request, while the employee remains across her eyeline near the right edge with enough office context to preserve their conversational axis.", "lighting_mood": "Use restrained office illumination with a cool desaturated palette and low-key contrast focused on 혜수's resolute profile.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": null, "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" ], "allowed_outlook_pairs": [ { "character_id": "C02", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "L13B01", "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", "describe the close subject and its immediate surroundings without relying on the absent background reference", "use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "applies_to_shots": [ 1, 4 ], "character_name": "", "confidence": "high", "creative_decisions": [], "description": "The scene remains inside the mart general affairs office, presented as a CCTV monitoring workspace. Maintain recognizable office context around the monitor-viewing and employee conversation areas without changing the established location.", "element_id": "cedar_harbor", "element_scope": "full", "element_type": "environment_state", "source_facts": [ "The scene heading identifies the location as the interior of the mart general affairs office.", "Shot 1 places the characters inside the mart general affairs office and CCTV control space.", "Shot 4 retains enough office context to preserve the conversational axis." ], "visual_inferences": [] }, { "applies_to_shots": [ 1, 4 ], "character_name": "", "confidence": "high", "creative_decisions": [], "description": "The office remains dimly and evenly restrained, with a cool, desaturated palette and low-key contrast. Do not introduce conspicuous additional light sources or shift the room toward warm, bright illumination.", "element_id": "moss_compass", "element_scope": "full", "element_type": "environment_state", "source_facts": [ "Shot 1 specifies subdued, cool-desaturated office lighting and instructs against inventing additional light sources.", "Shot 4 specifies restrained office illumination with a cool desaturated palette and low-key contrast." ], "visual_inferences": [] } ], "previous_shot_refs": [ { "scene_index": 20, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C02O02", "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" ] }, "render_contracts": [], "_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": "742bd9bdf24dd708" }, { "scene_index": 21, "heading": "S#21. 경찰서 안 / D", "beat_title": "충격적인 사건 현장 사진", "representative_moment": "수리영이 책상에 놓인 끔찍한 사건 현장 사진들을 보며 얼어붙은 순간", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "high three-quarter over-the-shoulder angle, neutral cool desaturated lighting", "t2i_prompt": "Photorealistic cinematic still. C01O01 in a casual hooded jacket, a young Korean woman with an oval face and long black hair tied back, sits hunched over in the lower-right foreground, looking toward the scattered P05 crime-scene photographs in the middle-center midground. Captured from a high three-quarter over-the-shoulder angle beside the desk, her upper body is bent forward, her posture rigid and completely frozen in shock, eyes noticeably widened in horror as she stares fixedly downward. The P05 photographs are spread across the central desk area, their image-bearing fronts facing upward toward the camera, displaying a reproduced deceased generic figure with severe neck and shoulder wounds. The police office environment is bathed in neutral, cool, desaturated illumination with restrained contrast, creating a stark, clinical atmosphere over the high-angle plane of the desk and the empty space beyond.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "혜수 책상위에는 살인사건 확대 사진들이 어지럽게 널려있다", "넘겨보며 놀라는 수리영", "shot description: over_shoulder, hunched over photographs, gaze=down" ], "visual_inferences": [ "fixed_element suriyoung_shocked_state: posture rigid and completely frozen in shock", "fixed_element office_cool_lighting: neutral, cool, desaturated illumination" ], "creative_decisions": [ "high three-quarter over-the-shoulder framing to simultaneously show C01O01's horrified reaction and the P05 photographs on the desk", "cool, clinical lighting to emphasize the grim nature of the murder evidence" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002" ], "reference_phrase_kinds": [ "character", "background" ], "owned_object_usage": [ { "owned_token": "cabinet", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "ceiling light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "desk", "usage_kind": "anchor", "source_phrase": "beside the desk, across the central desk area" }, { "owned_token": "document", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "keyboard", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "monitor", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "mouse", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "partition", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "photograph", "usage_kind": "redraw", "source_phrase": "scattered P05 crime-scene photographs" }, { "owned_token": "printer", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "telephone", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "e4008db5441ae034", "owned_hash": "530878202bc1e018", "camera_direction_hash": "00139f27f4cedd9e", "owned_usage_hash": "0ad1ef388723826b", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "desk", "violating_phrase": "beside the desk, across the central desk area, plane of the desk", "reason": "The desk is referred to solely as a spatial anchor and surface for the character's pose and placement of items, matching the producer's anchor declaration.", "verdict": "anchor_reference" }, { "owned_object": "photograph", "violating_phrase": "scattered P05 crime-scene photographs in the middle-center midground ... displaying a reproduced deceased generic figure with severe neck and shoulder wounds", "reason": "The prompt details the inner content of the photographs, which is structurally required by the shot_intent describing the character looking at '끔찍한 사건 현장 사진들' (gruesome crime scene photographs), qualifying for a visual state/content exception.", "verdict": "allowed_visual_state_change" } ] } } ], "scene_type": "normal", "dependent_scene_index": 16, "dependency_reason": "경찰서 사무실 공간 연속성 참조", "_shot_index": 3, "visible_entities": [ "L12", "C01", "P05" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 21, "shot_index": 3 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From the high three-quarter position beside 혜수's desk, continue the dolly-in until the camera hovers just above 수리영's seated shoulder. Arrange her bent upper body in the lower side of frame and spread the crime-scene photographs across the central desk area, letting her downward gaze lead directly into their disturbing details.", "lighting_mood": "Use neutral office illumination in a cool desaturated grade, with restrained contrast that keeps the printed evidence legible without stylization.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "looks_to_anchor", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "midground", "gesture_action": "none", "gesture_target_label": "", "label": "scattered crime-scene photographs", "screen_zone": "middle_center", "target_id": "", "target_kind": "background" }, { "constraint_id": "fsc_002", "depth_plane": "foreground", "gesture_action": "looks_toward", "gesture_target_label": "scattered crime-scene photographs", "label": "수리영", "screen_zone": "lower_right", "target_id": "C01", "target_kind": "character" } ] }, "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": [ "L12", "C01", "P05" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L12B02", "reference_usage": "exact_background", "owned_objects": [ "cabinet", "ceiling light", "chair", "desk", "document", "door", "keyboard", "monitor", "mouse", "partition", "photograph", "printer", "telephone", "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "suriyoung_shocked_state", "element_type": "character_state", "character_name": "수리영", "description": "A young Korean woman remains in a seated position with her upper body slightly hunched forward, her posture rigid and completely frozen in shock. Her eyes are noticeably widened in horror, and her intense gaze is locked fixedly downward. She maintains this paralyzed, overwhelmed physical state without moving.", "applies_to_shots": [ 3, 5 ], "element_scope": "full", "source_facts": [ "사무실에 홀로 앉아있는 수리영", "넘겨보며 놀라는 수리영", "사색이 된 수리영" ], "visual_inferences": [ "She is physically paralyzed by the gruesome nature of the evidence, maintaining a stiff and hunched posture across the shots." ], "creative_decisions": [ "Her downward gaze and wide-eyed frozen expression are emphasized to strongly convey her internal trauma upon seeing the photographs." ], "confidence": "high" }, { "element_id": "office_cool_lighting", "element_type": "environment_state", "character_name": "", "description": "The police office environment is bathed in neutral, cool, and desaturated illumination. The ambient lighting maintains a restrained and modest contrast, creating a stark, clinical, and serious atmosphere across the space. There is no heavy stylization or warm lighting present.", "applies_to_shots": [ 3, 5, 8 ], "element_scope": "full", "source_facts": [ "경찰서 안 / D" ], "visual_inferences": [ "The daytime office uses bright but stark fluorescent lighting, typical of a busy, pragmatic police station setting." ], "creative_decisions": [ "A cool, desaturated color grade with restrained contrast is applied consistently to enhance the grim tone of the murder investigation and the disturbing evidence." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 16, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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": "background", "id": "L12B02", "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" ] }, "render_contracts": [], "_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": "212c6cd7ad59aa23" }, { "scene_index": 21, "heading": "S#21. 경찰서 안 / D", "beat_title": "살인사건 사진을 확인하고 충격에 빠지는 수리영", "representative_moment": "사색이 되어 사건 현장 사진을 움켜쥔 채 굳어있는 C01의 얼굴 클로즈업.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "seated eye level tight three-quarter facial close-up with cool, desaturated lighting", "t2i_prompt": "Photorealistic cinematic still. C01O01 in a casual hooded jumper, a young Korean woman, is positioned in the middle-center foreground. Captured at seated eye level in a tight three-quarter frame from the right, her upper body is slightly hunched forward, posture rigid and completely frozen in shock. Her face and horror-widened eyes are sharply resolved, her intense gaze locked fixedly downward. Her hands gripping a photograph enter along the lower frame edge, the image-bearing front of the photograph partly turned toward her and only partially visible to the camera. The light is cool, desaturated, and restrained with modest facial contrast to hold the rigidity of her expression. [L12B02: The open police office space remains in soft focus behind her, bathed in stark, clinical, and serious ambient lighting.]", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "넘겨보며 놀라는 수리영", "사색이 된 수리영", "사진을 쥔 채 사색이 되어 두 눈을 크게 뜬 수리영의 굳은 얼굴" ], "visual_inferences": [ "fixed_element suriyoung_shocked_state: upper body slightly hunched forward, posture rigid and completely frozen in shock, intense gaze locked fixedly downward", "fixed_element office_cool_lighting: neutral, cool, and desaturated illumination, clinical and serious atmosphere" ], "creative_decisions": [ "Tight close-up framing at seated eye level to isolate the character's facial trauma", "The photograph only enters the lower frame edge to prioritize her facial reaction over the gruesome details", "Soft out-of-focus background to contrast her sharply resolved shocked expression against the empty office space" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001" ], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [ { "owned_token": "cabinet", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "ceiling light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "desk", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "document", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "keyboard", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "monitor", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "mouse", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "partition", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "photograph", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "printer", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "telephone", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "7ab84203afb307f2", "owned_hash": "530878202bc1e018", "camera_direction_hash": "cf88606a7279ad2e", "owned_usage_hash": "9d1d74a6d014f7f0", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 21, "dependency_reason": "Same continuous police office location, maintaining the consistent cool and clinical lighting atmosphere established in previous shots.", "_shot_index": 5, "visible_entities": [ "C01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 21, "shot_index": 5 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "End the desk-side arc at 수리영's seated eye level in a tight three-quarter facial frame, keeping her rigid face and widened eyes sharply resolved. Her clenched photograph enters only along the lower edge as contextual evidence, while open office space remains soft behind her in the direction the camera will next track.", "lighting_mood": "Keep the light cool, desaturated, and restrained, using modest facial contrast to hold the rigidity of 수리영's shocked expression.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "primary_subject_isolation", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "foreground", "gesture_action": "none", "gesture_target_label": "", "label": "수리영", "screen_zone": "middle_center", "target_id": "C01", "target_kind": "character" } ] }, "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" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "L12B02", "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", "describe the close subject and its immediate surroundings without relying on the absent background reference", "use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "suriyoung_shocked_state", "element_type": "character_state", "character_name": "수리영", "description": "A young Korean woman remains in a seated position with her upper body slightly hunched forward, her posture rigid and completely frozen in shock. Her eyes are noticeably widened in horror, and her intense gaze is locked fixedly downward. She maintains this paralyzed, overwhelmed physical state without moving.", "applies_to_shots": [ 3, 5 ], "element_scope": "full", "source_facts": [ "사무실에 홀로 앉아있는 수리영", "넘겨보며 놀라는 수리영", "사색이 된 수리영" ], "visual_inferences": [ "She is physically paralyzed by the gruesome nature of the evidence, maintaining a stiff and hunched posture across the shots." ], "creative_decisions": [ "Her downward gaze and wide-eyed frozen expression are emphasized to strongly convey her internal trauma upon seeing the photographs." ], "confidence": "high" }, { "element_id": "office_cool_lighting", "element_type": "environment_state", "character_name": "", "description": "The police office environment is bathed in neutral, cool, and desaturated illumination. The ambient lighting maintains a restrained and modest contrast, creating a stark, clinical, and serious atmosphere across the space. There is no heavy stylization or warm lighting present.", "applies_to_shots": [ 3, 5, 8 ], "element_scope": "full", "source_facts": [ "경찰서 안 / D" ], "visual_inferences": [ "The daytime office uses bright but stark fluorescent lighting, typical of a busy, pragmatic police station setting." ], "creative_decisions": [ "A cool, desaturated color grade with restrained contrast is applied consistently to enhance the grim tone of the murder investigation and the disturbing evidence." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 21, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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" } ], "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" ] }, "render_contracts": [], "_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": "f08c4bf3288c3bce" }, { "scene_index": 21, "heading": "S#21. 경찰서 안 / D", "beat_title": "엄마 폰의 해상 위치추적 결과 확인", "representative_moment": "모니터 화면 중앙에 바다 위를 가리키는 붉은 점표시와 오프라인 상태 정보가 선명하게 떠 있다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Insert close-up on the monitor display, cool desaturated digital aesthetic with a selective crimson accent, subtle screen texture.", "t2i_prompt": "Photorealistic cinematic still. A monitor tracking-result display dominates the middle-center foreground, presenting a glowing digital maritime map. The active screen shows an offshore tracking result, marked by a sharp crimson red coordinate point that stands out against the cool, desaturated blue tones of the sea map. Identifying data fields including a name, numerical coordinates, a timestamp, GPS source, and an offline status indicator are crisply rendered on the interface with a slight digital pixel texture. A thin black bezel of the monitor tightly frames the display, with the edge of a desk faintly visible in the clinical, stark fluorescent office lighting beyond the screen.", "outfit_assignments": [], "source_facts": [ "해상 지도위에 좌표. 37.484400 N / 126. 072486 E", "Timestamp 2026-02-18 11:21:08", "Status: offline", "바다 위 특정 해상 좌표가 붉은 점으로 표시된 모니터 지도 화면" ], "visual_inferences": [ "office_cool_lighting: neutral, cool, and desaturated illumination creating a clinical atmosphere", "thin black bezel of the monitor framing the shot", "digital pixel texture to convey the through-device perception mode" ], "creative_decisions": [ "Insert framing to emphasize the crucial tracking information", "Selective crimson red against cool desaturated tones to heighten the ominous discovery" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001" ], "reference_phrase_kinds": [ "background" ], "owned_object_usage": [ { "owned_token": "cabinet", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "ceiling light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "desk", "usage_kind": "anchor", "source_phrase": "edge of a desk faintly visible" }, { "owned_token": "document", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "keyboard", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "monitor", "usage_kind": "anchor", "source_phrase": "A monitor tracking-result display" }, { "owned_token": "mouse", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "partition", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "photograph", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "printer", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "telephone", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "241e52e5c013feb3", "owned_hash": "530878202bc1e018", "camera_direction_hash": "9026058fa61b8e2f", "owned_usage_hash": "232914b568736232", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "desk", "violating_phrase": "edge of a desk faintly visible", "reason": "The prompt only references the desk's edge as a faint background element to establish composition, matching the producer's 'anchor' declaration.", "verdict": "anchor_reference" }, { "owned_object": "monitor", "violating_phrase": "A monitor tracking-result display dominates the middle-center foreground, presenting a glowing digital maritime map", "reason": "The prompt describes specific tracking data and coordinates displayed on the monitor screen, which is permitted under visual state change exception 1 as explicitly required by the shot_intent ('모니터 화면 중앙에 바다 위를 가리키는 붉은 점표시와 오프라인 상태 정보가 선명하게 떠 있다').", "verdict": "allowed_visual_state_change" } ] } } ], "scene_type": "normal", "dependent_scene_index": 21, "dependency_reason": "이전 샷(S21_Shot5)의 경찰서 사무실 배경 조명 및 공간 톤 유지", "_shot_index": 8, "visible_entities": [], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 21, "shot_index": 8 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "insert", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Terminate the frontal dolly-in square to 김형사의 monitor, with the maritime map and red coordinate marker spanning the frame while a thin monitor boundary remains visible. Center the offshore point and keep the identifying fields—강민숙's name, last-location coordinates, timestamp, source, and offline status—readable around it.", "lighting_mood": "Let the active display remain clear within a cool desaturated treatment, using the documented red coordinate point as the selective crimson accent.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "required_background_position", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "foreground", "gesture_action": "none", "gesture_target_label": "", "label": "monitor tracking-result display", "screen_zone": "middle_center", "target_id": "", "target_kind": "background" } ] }, "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": [], "allowed_outlook_pairs": [], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L12B02", "reference_usage": "exact_background", "owned_objects": [ "cabinet", "ceiling light", "chair", "desk", "document", "door", "keyboard", "monitor", "mouse", "partition", "photograph", "printer", "telephone", "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "office_cool_lighting", "element_type": "environment_state", "character_name": "", "description": "The police office environment is bathed in neutral, cool, and desaturated illumination. The ambient lighting maintains a restrained and modest contrast, creating a stark, clinical, and serious atmosphere across the space. There is no heavy stylization or warm lighting present.", "applies_to_shots": [ 3, 5, 8 ], "element_scope": "full", "source_facts": [ "경찰서 안 / D" ], "visual_inferences": [ "The daytime office uses bright but stark fluorescent lighting, typical of a busy, pragmatic police station setting." ], "creative_decisions": [ "A cool, desaturated color grade with restrained contrast is applied consistently to enhance the grim tone of the murder investigation and the disturbing evidence." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 21, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "background", "id": "L12B02", "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" ] }, "render_contracts": [], "_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": "063e9decd6336500" }, { "scene_index": 22, "heading": "S#22. 경찰서 밖 / D", "beat_title": "수리영의 다급한 출발", "representative_moment": "C01이 경찰서 앞 주차장에서 자전거 페달을 힘껏 밟으며 밖으로 내달리는 순간.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "waist-height observer view with full-body side profile", "t2i_prompt": "Photorealistic cinematic still. C01O01 in a casual hooded jumper, an Asian 21-year-old woman, occupies the middle-center midground, moving toward the right edge of the frame, with the front exterior of the pre-existing police station building dominating the upper-center background. The camera is positioned at waist height, framing the figure in a full-body right profile, seated on the saddle of the pre-existing bicycle. One bent knee rises while the opposite foot drives the descending pedal of the bicycle down. Her urgent face is seen in right profile, gazing into the distance, her long black hair flowing back in the direction opposite to motion. The scene is lit by natural ambient daylight graded cool and desaturated, casting restrained contrast over the pre-existing entrance door and the road surface.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "S#22. 경찰서 밖 / D 주차된 자전거를 타고 이내 달리는 수리영", "C01: profile_right, driving pedal downward, gaze=distant, state=alive", "Shot description: 경찰서 건물 앞, 자전거 안장에 앉은 채 한쪽 무릎을 굽혀 페달을 힘껏 밟아 내리는 중간 찰나, 다급한 표정의 수리영의 측면" ], "visual_inferences": [ "long black hair flowing back indicating forward motion", "waist-height camera capturing the full motion profile of the bicycle" ], "creative_decisions": [ "Placed C01 in the middle-center moving right to convey the DP's direction of travel", "Applied cool desaturated daylight for the exterior atmosphere and restrained contrast to highlight the action" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002" ], "reference_phrase_kinds": [ "character", "background" ], "owned_object_usage": [ { "owned_token": "AC", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "CCTV camera", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "bicycle", "usage_kind": "anchor", "source_phrase": "seated on the saddle of the pre-existing bicycle" }, { "owned_token": "bicycle rack", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "bollard", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "curb", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "drainpipe", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "entrance door", "usage_kind": "anchor", "source_phrase": "the pre-existing entrance door" }, { "owned_token": "flagpole", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "parking line", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "planter", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "police station", "usage_kind": "anchor", "source_phrase": "the front exterior of the pre-existing police station building" }, { "owned_token": "road", "usage_kind": "anchor", "source_phrase": "the road surface" }, { "owned_token": "sign", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "f653c894e4b01ffb", "owned_hash": "31e19cf50f36c010", "camera_direction_hash": "2754e80e1553d435", "owned_usage_hash": "1e212aaceb0c6397", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "bicycle", "violating_phrase": "seated on the saddle of the pre-existing bicycle. One bent knee rises while the opposite foot drives the descending pedal of the bicycle down", "reason": "The character physically interacts with the bicycle to pedal it, specifying contact geometry with the saddle and pedals while respecting the pre-existing object.", "verdict": "allowed_prop_contact" }, { "owned_object": "entrance door", "violating_phrase": "the pre-existing entrance door", "reason": "The entrance door is mentioned solely as a reference for casting ambient light and shadow, aligning with the anchor declaration.", "verdict": "anchor_reference" }, { "owned_object": "police station", "violating_phrase": "the front exterior of the pre-existing police station building", "reason": "The police station is referenced strictly as a background composition anchor to set the scene, requiring no redesign.", "verdict": "anchor_reference" }, { "owned_object": "road", "violating_phrase": "the road surface", "reason": "The road surface is referenced purely for lighting/contrast casting, acting as an anchor in the scene.", "verdict": "anchor_reference" } ] } } ], "scene_type": "normal", "dependent_scene_index": 21, "dependency_reason": "경찰서 외부 배경(S21의 exact_background) 및 공간의 연속성 유지", "_shot_index": 1, "visible_entities": [ "C01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 22, "shot_index": 1 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "wide", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At waist height and perpendicular to the bicycle, hold the inherited side axis on the exact launch instant before the lateral track begins. Frame 수리영 and the bicycle in a full-body side view as one bent knee rises and the opposite foot drives its pedal down, her urgent face angled toward the direction of travel.", "lighting_mood": "Use natural ambient daylight appropriate to the exterior, graded cool and desaturated with restrained contrast and no added atmospheric effects.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "movement_direction", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "background", "gesture_action": "none", "gesture_target_label": "", "label": "police station building", "screen_zone": "upper_center", "target_id": "", "target_kind": "background" }, { "constraint_id": "fsc_002", "depth_plane": "midground", "gesture_action": "moves_toward", "gesture_target_label": "right edge of frame", "label": "수리영", "screen_zone": "middle_center", "target_id": "C01", "target_kind": "character" } ] }, "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" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L12B01", "reference_usage": "exact_background", "owned_objects": [ "AC", "CCTV camera", "bicycle", "bicycle rack", "bollard", "curb", "drainpipe", "entrance door", "flagpole", "parking line", "planter", "police station", "road", "sign", "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [ { "scene_index": 21, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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": "background", "id": "L12B01", "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" ] }, "render_contracts": [], "_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": "f54b2bced9f901a2" }, { "scene_index": 23, "heading": "S#23. 어느 PC방 / 실내", "beat_title": "좌표 확인", "representative_moment": "PC방 모니터 화면에 해상 좌표와 붉은 점이 표시된 지도.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "perpendicular insert framing, cool desaturated PC-room lighting", "t2i_prompt": "Photorealistic cinematic still. A PC monitor maritime map display is positioned in the middle-center foreground, facing the camera nearly square-on. A slim dark screen border remains visible around the extreme edges, serving as device context. The digital interface glows with cool, desaturated blue and gray topography, displaying crisp coordinate text. A single distinct red point marks an offshore location, standing out as a sharp crimson accent against the muted maritime map. Subtle digital pixel texture and faint ambient glare across the glass surface ground the device in a dark room setting, with no characters introduced into the frame.", "outfit_assignments": [], "source_facts": [ "S#23 action: '좌표를 찍고 위치를 확인하고'", "shot description: 해상 좌표가 붉은 점으로 뚜렷하게 표시된 PC방 모니터 화면 클로즈업" ], "visual_inferences": [ "maritime digital map interface showing topography and text", "dark PC-room ambient light reflecting subtly on the monitor bezel" ], "creative_decisions": [ "perpendicular insert framing focusing purely on the screen content to emphasize the coordinates", "cool desaturated palette contrasting with the distinct crimson red marker for visual impact", "inclusion of a slim screen border to establish the physical monitor context without introducing the character" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001" ], "reference_phrase_kinds": [], "owned_object_usage": [], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "5bd6e3272eb4c1bd", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "ea8a82409c35a5f9", "owned_usage_hash": "4f53cda18c2baa0c", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "", "_shot_index": 2, "visible_entities": [], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 23, "shot_index": 2 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "insert", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Hold very close and almost perpendicular to the PC monitor so the maritime map, coordinate text, and distinct red point are geometrically undistorted and immediately readable. Retain a slim screen border as device context, centering the offshore marker rather than introducing 수리영 into the frame.", "lighting_mood": "Keep the PC-room image cool and desaturated while preserving the documented red map point as a precise crimson accent.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "required_background_position", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "foreground", "gesture_action": "none", "gesture_target_label": "", "label": "PC monitor maritime map display", "screen_zone": "middle_center", "target_id": "", "target_kind": "background" } ] }, "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": [], "allowed_outlook_pairs": [], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "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" ] }, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [], "forbidden_refs": [], "readiness_policy": "not_applicable", "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" ] }, "render_contracts": [], "_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": "5e2b4aa322afd0b3" }, { "scene_index": 24, "heading": "S#24. 해안가 / 소나기", "beat_title": "빗속의 절박한 질주", "representative_moment": "거센 빗줄기를 맞으며 자전거 페달을 강하게 밟아 해안도로를 내달리는 수리영(C01)의 모습.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "wide lateral tracking composition at chest-to-waist height, cool desaturated storm light with lightning highlights", "t2i_prompt": "Photorealistic cinematic still. Seen from a low chest-to-waist tracking angle, C01O01 in a casual hooded jumper, a Korean young woman, occupies the middle-center midground in a wide lateral three-quarter composition, moving toward the frame-right coastal route. Her torso is pitched forward at three-quarter angle from the right, eyes fixed off-screen with intense desperation. Captured in mid-pedal, she drives the pedal forcefully downward, her weighted foot clear in the lower center. Hands firmly grip the handlebar, with the bicycle's main frame and front wheel fully visible below her body. Wet hair and jacket hem stream backward toward the left edge, opposite to the forward motion. Heavy rain cuts across the frame as visible diagonal texture, illuminated by cool desaturated storm light and brief lightning-driven highlights catching her wet skin and the metallic handlebars. Gloomy coastal surroundings and dark storm clouds recede into the upper background behind the rider.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "S#24 action: '자전거를 타고 빗속을 달리는 수리영.'", "shot description: 세차게 쏟아지는 소나기 속에서 흠뻑 젖은 채, 한쪽 발로 자전거 페달을 강하게 내리누르는 몸이 앞으로 쏠린 수리영의 전신" ], "visual_inferences": [ "forward-pitched torso mid-action freeze to convey dynamic pedaling momentum", "hair and jacket trailing backward opposite to the direction of motion", "lightning reflections on wet skin and bicycle frame to emphasize the storm" ], "creative_decisions": [ "chest-to-waist lateral tracking angle to capture both the downward pedal force and the full body traveling horizontally", "cool desaturated palette with stark lightning contrast to highlight the dramatic urgency" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001" ], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "3d64c1ce142cad16", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "831db0901dd01503", "owned_usage_hash": "4f53cda18c2baa0c", "validator": "scene_detail_owned_objects.v1", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "N/A", "_shot_index": 2, "visible_entities": [ "L16", "C01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 24, "shot_index": 2 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "wide", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Track laterally alongside 수리영 at a low chest-to-waist height, holding her full body and bicycle in a wide lateral three-quarter composition as she drives toward frame right. Her forward-pitched torso occupies the center-left while the forcefully lowered pedal and weighted foot remain clear in the lower center, with the coastal surroundings receding behind her.", "lighting_mood": "Cool desaturated storm light with restrained contrast and brief lightning-driven highlights through the heavy rain.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "movement_direction", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "midground", "gesture_action": "moves_toward", "gesture_target_label": "frame-right coastal route", "label": "수리영", "screen_zone": "middle_center", "target_id": "C01", "target_kind": "character" } ] }, "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": [ "L16", "C01" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "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" ] }, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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" } ], "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" ] }, "render_contracts": [], "_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": "c552d331b48aa1d0" }, { "scene_index": 25, "heading": "S#25. 옥탑방 안 / 해질 무렵", "beat_title": "엉망이 된 방과 충격", "representative_moment": "어질러진 가재도구 사이에서 충격에 빠진 채 열린 방문 쪽을 바라보는 C01.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "observer point of view at chest height, slightly off-center medium composition, cool desaturated dusk", "t2i_prompt": "Photorealistic cinematic still. At chest height, waist-up of C01O01 in a casual hooded jumper, an East Asian young woman with an oval face and long black tied hair, standing slightly off-center. She is positioned at three-quarter angle from the left, leaning back in shock, eyes looking off-screen toward the open doorway obliquely facing the camera. She is surrounded by scattered household goods and a fallen chair with its side facing the camera. The open living-room window holds an empty void behind and beside her. Cool, subdued dusk ambience fills the apartment in muted, desaturated tones, with restrained low-key contrast and selective deep-crimson details, maintaining an unobstructed view across the room.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "S#25 scene text: 옥탑방 안 / 해질 무렵. 집안에서 뭔가를 찾기 위해 가재도구를 온통 헤집은 것으로 보인다.", "Shot 2 description: 온갖 가재도구가 쏟아져 엉망이 된 거실 한가운데 선 수리영의 충격받은 상체.", "DP directions: observer POV, chest height directly across the disrupted living room, waist-up slightly off-center medium composition." ], "visual_inferences": [ "fixed_element amber_lattice: bedroom doorway remains open, maintaining an unobstructed view.", "fixed_element silent_meridian: subdued dusk ambience in muted desaturated tones, restrained contrast with selective deep-crimson accents.", "gaze=off_screen translates to looking toward the open doorway obliquely facing the camera." ], "creative_decisions": [ "Placed the fallen chair and scattered household goods prominently around the lower frame to emphasize the visual chaos.", "Utilized the open living-room window to create spatial depth behind the character while keeping the lighting moody and dim." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [], "reference_phrase_kinds": [ "character", "background" ], "owned_object_usage": [ { "owned_token": "cabinet", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "chair", "usage_kind": "anchor", "source_phrase": "a fallen chair" }, { "owned_token": "curtain", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "dining table", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "drawer", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "entrance door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "floor lamp", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "footprint", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "household goods", "usage_kind": "anchor", "source_phrase": "scattered household goods" }, { "owned_token": "living-room window", "usage_kind": "anchor", "source_phrase": "open living-room window" }, { "owned_token": "sofa", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "d75b24566f6fdce0", "owned_hash": "108e3b65bc2770f6", "camera_direction_hash": "7e5bebcf49ac557a", "owned_usage_hash": "1edc23de3c2ef3a9", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "chair", "violating_phrase": "a fallen chair with its side facing the camera", "reason": "The prompt uses the fallen chair to establish the scene's layout and composition without instructing to redraw or modify the background object, aligning with the producer's anchor classification.", "verdict": "anchor_reference" }, { "owned_object": "household goods", "violating_phrase": "scattered household goods", "reason": "The prompt mentions the scattered household goods as spatial context surrounding the character, acting as an anchor in accordance with the producer's declaration.", "verdict": "anchor_reference" }, { "owned_object": "living-room window", "violating_phrase": "The open living-room window", "reason": "The prompt references the open living-room window as a compositional anchor to define the empty space behind the character, matching the producer's anchor classification.", "verdict": "anchor_reference" } ] } } ], "scene_type": "normal", "dependent_scene_index": 18, "dependency_reason": "동일한 옥탑방 배경을 공유하며, 앞선 씬에서 설정된 공간적 연속성(이전 샷의 위치와 레이아웃)을 exact_background로 유지해야 함", "_shot_index": 2, "visible_entities": [ "L05", "C01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 25, "shot_index": 2 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Begin the reaction stage at chest height directly across the disrupted living room, framing 수리영 from the waist up in a slightly off-center medium composition. Keep the fallen chairs and scattered household items around her while her compressed breathing and tense glance toward the open bedroom doorway initiate the coming pan.", "lighting_mood": "Dim dusk ambience is rendered in cool desaturated tones, with restrained contrast and no added motivated source.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": null, "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": [ "L05", "C01" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L05B04", "reference_usage": "exact_background", "owned_objects": [ "cabinet", "chair", "curtain", "dining table", "drawer", "entrance door", "floor lamp", "footprint", "household goods", "living-room window", "sofa" ], "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "applies_to_shots": [ 2, 3 ], "character_name": "", "confidence": "high", "creative_decisions": [], "description": "The bedroom doorway remains open, maintaining an unobstructed view from the disrupted living room into the bedroom. Its position and open state remain unchanged across both shots.", "element_id": "amber_lattice", "element_scope": "full", "element_type": "environment_state", "source_facts": [ "Shot 2 describes a tense glance toward the open bedroom doorway.", "Shot 3 moves through the open doorway toward the bedroom wall.", "The scene text states that her room is open." ], "visual_inferences": [] }, { "applies_to_shots": [ 2, 3, 5 ], "character_name": "", "confidence": "high", "creative_decisions": [], "description": "Cool, subdued dusk ambience fills the rooftop apartment in muted, desaturated tones. Contrast remains restrained and low-key, with dark red and deep-crimson details serving as selective accents rather than bright highlights.", "element_id": "silent_meridian", "element_scope": "full", "element_type": "environment_state", "source_facts": [ "The scene takes place inside the rooftop apartment at dusk.", "Shot 2 specifies dim dusk ambience in cool desaturated tones with restrained contrast.", "Shot 3 specifies cool, subdued dusk ambience and low-key restraint.", "Shot 5 specifies muted cool tones around the photograph." ], "visual_inferences": [] } ], "previous_shot_refs": [ { "scene_index": 18, "shot_index": 8, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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": "background", "id": "L05B04", "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" ] }, "render_contracts": [], "_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": "5ea5699e6e684771" }, { "scene_index": 25, "heading": "S#25. 옥탑방 안 / 해질 무렵", "beat_title": "붉은 원과 사진의 발견", "representative_moment": "A subjective view through the open doorway reveals a red circle painted on the bedroom wall with the P03 photograph fixed at its center.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Subjective push-in framing at eye level from the doorway, keeping the photograph centered and proportional, with low-key restrained dusk lighting.", "t2i_prompt": "Photorealistic cinematic still. The open bedroom doorway in the middle-center foreground frames the wall with the P02 wall marking and the centered P03 photograph in the middle-center background. [L05: Cool, subdued dusk ambience fills the rooftop apartment space in muted, desaturated tones, the interior kept in low-key restraint without revealing the surrounding furniture.] On the camera-facing wall, P02, a vivid deep-crimson wall marking, dominates the view. Anchored at the exact center of P02 is the P03 photograph, remaining a small handheld snapshot size relative to the wall, with its front facing the camera. Contrast remains restrained, with the red circle serving as a selective accent against the dim dusk.", "outfit_assignments": [], "source_facts": [ "Shot 3 describes a view through the open doorway toward the bedroom wall.", "A vivid red circle is painted on the wall with a photograph fixed at its center.", "The scene text identifies the photograph as showing a circa-1970s-or-1980s rural restaurant with a circular rooftop marker and two young girls." ], "visual_inferences": [ "fixed_element amber_lattice: The bedroom doorway remains open, maintaining an unobstructed view into the room.", "fixed_element silent_meridian: Cool, subdued dusk ambience in muted, desaturated tones dominates the space.", "fixed_element cedar_token: The photograph bears persistent dark red stage-paint marks across its surface.", "Medium framing captures the central occult composition from a subjective POV, excluding Suri-young from the frame." ], "creative_decisions": [ "Used the doorway edges as a natural foreground frame in the middle-center to anchor the subjective push-in.", "Maintained low-key contrast to ensure the deep-crimson wall marking and paint act as the primary visual accents." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002" ], "reference_phrase_kinds": [ "background", "prop" ], "owned_object_usage": [ { "owned_token": "bed", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "bedding", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "curtain", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "photograph", "usage_kind": "anchor", "source_phrase": "Anchored at the exact center of P02 is the P03 photograph, remaining a small handheld snapshot size relative to the wall, with its front facing the camera." }, { "owned_token": "pin", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wall marking", "usage_kind": "anchor", "source_phrase": "On the camera-facing wall, P02, a vivid deep-crimson wall marking, dominates the view" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "8fd796b1737f3ebc", "owned_hash": "a08bc85f4434aeb1", "camera_direction_hash": "1251567006391320", "owned_usage_hash": "66672cf8b67ab269", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "wall marking", "violating_phrase": "a vivid deep-crimson wall marking, dominates the view", "reason": "The prompt references the wall marking to establish composition and describes its appearance as an existing anchor, aligning with the producer's anchor declaration.", "verdict": "anchor_reference" }, { "owned_object": "photograph", "violating_phrase": "P03 photograph, remaining a small handheld snapshot size relative to the wall", "reason": "The prompt describes the photograph's position and scale relative to the wall as an anchor, matching the producer's declaration with no redraw intent.", "verdict": "anchor_reference" } ] } } ], "scene_type": "normal", "dependent_scene_index": 12, "dependency_reason": "The P03 photograph and its specific visual contents visually tie back to S12_Shot9, requiring consistency for the vintage restaurant and the young girls depicted.", "_shot_index": 3, "visible_entities": [ "L05", "P02", "P03" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 25, "shot_index": 3 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From 수리영's direct line of sight at eye height, begin a measured push through the open doorway toward the bedroom wall. The doorway edges frame the wall, placing the crimson circle in the middle field and the blood-stained photograph at its exact visual center without letting the photograph exceed a natural fraction of the frame.", "lighting_mood": "Cool, subdued dusk ambience holds the room in low-key restraint while the deep-crimson circle and blood remain the selective accents.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "required_background_position", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "foreground", "gesture_action": "none", "gesture_target_label": "", "label": "open bedroom doorway", "screen_zone": "middle_center", "target_id": "", "target_kind": "background" }, { "constraint_id": "fsc_002", "depth_plane": "background", "gesture_action": "none", "gesture_target_label": "", "label": "wall with red circle and centered photograph", "screen_zone": "middle_center", "target_id": "", "target_kind": "background" } ] }, "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": [ "L05", "P02", "P03" ], "allowed_outlook_pairs": [], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L05B03", "reference_usage": "exact_background", "owned_objects": [ "bed", "bedding", "curtain", "photograph", "pin", "wall marking", "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "applies_to_shots": [ 2, 3 ], "character_name": "", "confidence": "high", "creative_decisions": [], "description": "The bedroom doorway remains open, maintaining an unobstructed view from the disrupted living room into the bedroom. Its position and open state remain unchanged across both shots.", "element_id": "amber_lattice", "element_scope": "full", "element_type": "environment_state", "source_facts": [ "Shot 2 describes a tense glance toward the open bedroom doorway.", "Shot 3 moves through the open doorway toward the bedroom wall.", "The scene text states that her room is open." ], "visual_inferences": [] }, { "applies_to_shots": [ 2, 3, 5 ], "character_name": "", "confidence": "high", "creative_decisions": [], "description": "Cool, subdued dusk ambience fills the rooftop apartment in muted, desaturated tones. Contrast remains restrained and low-key, with dark red and deep-crimson details serving as selective accents rather than bright highlights.", "element_id": "silent_meridian", "element_scope": "full", "element_type": "environment_state", "source_facts": [ "The scene takes place inside the rooftop apartment at dusk.", "Shot 2 specifies dim dusk ambience in cool desaturated tones with restrained contrast.", "Shot 3 specifies cool, subdued dusk ambience and low-key restraint.", "Shot 5 specifies muted cool tones around the photograph." ], "visual_inferences": [] }, { "applies_to_shots": [ 3, 5 ], "character_name": "", "confidence": "high", "creative_decisions": [], "description": "An aged photograph prop bears persistent dark red stage-paint marks across its surface. The photograph shows a rural restaurant from around the 1970s or 1980s, a circular marker mounted on the roof, and two approximately twelve-year-old girls seated together in front of the building.", "element_id": "cedar_token", "element_scope": "full", "element_type": "persistent_prop", "source_facts": [ "Shot 3 shows a dark-red-stained photograph fixed at the exact center of the crimson wall circle.", "Shot 5 shows the same aged, red-paint-marked photograph in detail.", "The scene text identifies the image as a circa-1970s-or-1980s rural restaurant with a circular rooftop marker and two twelve-year-old girls seated in front." ], "visual_inferences": [] } ], "previous_shot_refs": [ { "scene_index": 12, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "background", "id": "L05B03", "policy": "required" }, { "kind": "prop", "id": "P03", "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" ] }, "render_contracts": [ { "contract_id": "rc_001", "scope": { "scene_index": 25, "shot_indices": [ 3 ], "duration": "single_shot" }, "targets": [ { "entity_id": "P03", "role": "visual_target" } ], "requirements": [ { "dimension": "visual_identity", "operation": "preserve", "strength": "required", "reference_policy": "use_entity_reference" } ] } ], "_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": "e2833f63611a6cfb" }, { "scene_index": 25, "heading": "S#25. 옥탑방 안 / 해질 무렵", "beat_title": "Realizing the truth behind the bloody photo", "representative_moment": "A close-up of P03, an aged and bloody photograph showing a 1970s rural restaurant, held by fingertips.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Close framing at eye level, sharp focus on the photograph against a softly blurred background, subdued dusk tones", "t2i_prompt": "Photorealistic cinematic still. In close framing at eye level, fingertips gently hold P03, an aged vintage photograph, keeping its front face occupying the lower-center frame. The image-bearing front of P03 faces the camera, clearly displaying a 1970s rural restaurant with a circular marker mounted on the roof. In the photograph, two Asian 12-year-old girls are seated together in front of the building; one Asian 12-year-old girl is in three-quarter right profile gazing into the distance, while the second Asian 12-year-old girl sits close beside her in three-quarter left profile, also gazing into the distance. Persistent dark red stage-paint marks stain the surface of P03. [: A narrow out-of-focus trace of the room behind is visible as a scale reference, filled with cool, subdued dusk ambience in muted desaturated tones]. Contrast is restrained and low-key, allowing the deep-crimson blood marks on the photograph to serve as a selective accent against the dim dusk fill.", "outfit_assignments": [], "source_facts": [ "Shot 5 specifies the aged, red-paint-marked photograph in detail.", "The photograph shows a 1970s/1980s rural restaurant, a rooftop circular marker, and two twelve-year-old girls seated in front.", "The inward move ends on the photograph held by her fingertips." ], "visual_inferences": [ "fixed_element silent_meridian: Cool, subdued dusk ambience fills the space in muted, desaturated tones with restrained contrast.", "fixed_element cedar_token: An aged photograph prop (P03) bearing persistent dark red stage-paint marks.", "The close framing isolates the photograph and holding fingertips, leaving the room as a soft, out-of-focus background trace." ], "creative_decisions": [ "Used tight close-up framing to anchor the viewer's attention entirely on the visual evidence inside the photograph and the blood marks.", "Applied cool, muted dusk tones with a deep-crimson accent to heighten the eerie, occult thriller mood." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [], "reference_phrase_kinds": [ "prop" ], "owned_object_usage": [ { "owned_token": "bed", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "bedding", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "curtain", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "photograph", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "pin", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wall marking", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "a150fd187d257b23", "owned_hash": "a08bc85f4434aeb1", "camera_direction_hash": "a3021c253a661b30", "owned_usage_hash": "cdf83c41427d4530", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 25, "dependency_reason": "zoom_in_detail on the photograph established in the previous shot", "_shot_index": 5, "visible_entities": [ "P03" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 25, "shot_index": 5 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At 수리영's eye height, end the inward move on the photograph held near her viewing line, keeping its front face under two-fifths of the frame with her fingertips and a narrow trace of the room preserved as scale references. The old restaurant, rooftop circular marker, and two seated girls remain readable within the image, preparing a graphic cut into the represented past.", "lighting_mood": "Muted cool tones surround the aged photograph, with its blood marks held as a restrained deep-crimson accent.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": null, "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": [ "P03" ], "allowed_outlook_pairs": [], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "L05B03", "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", "describe the close subject and its immediate surroundings without relying on the absent background reference", "use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "applies_to_shots": [ 2, 3, 5 ], "character_name": "", "confidence": "high", "creative_decisions": [], "description": "Cool, subdued dusk ambience fills the rooftop apartment in muted, desaturated tones. Contrast remains restrained and low-key, with dark red and deep-crimson details serving as selective accents rather than bright highlights.", "element_id": "silent_meridian", "element_scope": "full", "element_type": "environment_state", "source_facts": [ "The scene takes place inside the rooftop apartment at dusk.", "Shot 2 specifies dim dusk ambience in cool desaturated tones with restrained contrast.", "Shot 3 specifies cool, subdued dusk ambience and low-key restraint.", "Shot 5 specifies muted cool tones around the photograph." ], "visual_inferences": [] }, { "applies_to_shots": [ 3, 5 ], "character_name": "", "confidence": "high", "creative_decisions": [], "description": "An aged photograph prop bears persistent dark red stage-paint marks across its surface. The photograph shows a rural restaurant from around the 1970s or 1980s, a circular marker mounted on the roof, and two approximately twelve-year-old girls seated together in front of the building.", "element_id": "cedar_token", "element_scope": "full", "element_type": "persistent_prop", "source_facts": [ "Shot 3 shows a dark-red-stained photograph fixed at the exact center of the crimson wall circle.", "Shot 5 shows the same aged, red-paint-marked photograph in detail.", "The scene text identifies the image as a circa-1970s-or-1980s rural restaurant with a circular rooftop marker and two twelve-year-old girls seated in front." ], "visual_inferences": [] } ], "previous_shot_refs": [ { "scene_index": 25, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "prop", "id": "P03", "policy": "required" } ], "forbidden_refs": [ { "kind": "background", "reason": "close framing skips chain_bg reference" } ], "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" ] }, "render_contracts": [ { "contract_id": "rc_001", "scope": { "scene_index": 25, "shot_indices": [ 5 ], "duration": "single_shot" }, "targets": [ { "entity_id": "P03", "role": "visual_target" } ], "requirements": [ { "dimension": "visual_identity", "operation": "preserve", "strength": "required", "reference_policy": "use_entity_reference" } ] } ], "_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": "dcedcf5de1d05038" }, { "scene_index": 26, "heading": "S#26. 포구 / 해질 무렵", "beat_title": "고립된 포구와 부재중 전화", "representative_moment": "C01's hand cradling P01, displaying missed calls in the foggy dusk.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Close-up hand-height framing, cool desaturated dusk lighting with heavy fog diffusion", "t2i_prompt": "Photorealistic cinematic still. Held at hand height beside the seated figure, close framing focuses on P01, a modern handheld smartphone occupying less than two-fifths of the frame. C01 in a casual hooded jumper, a young East Asian woman, cradles P01 in her hand, her gripping fingers, a portion of her seated torso, and subtly damp hair providing scale. She is looking away with gaze down, her posture alive but exhausted. The active front display of P01 faces the camera, brightly illuminated to show caller identification for Hye-su with multiple missed-call notifications. Behind the hand and device, [: a rocky coastal edge enveloped in thick, dense fog during the cool desaturated dusk, keeping the immediate background subdued and out of focus while the phone screen provides local brightness].", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "shot description: 수리영의 손안에서 혜수의 이름과 부재중 전화 알림이 떠오른 휴대전화 액정 화면 클로즈업", "scene text: 얼핏 보니 혜수이모다. 벌써 여러통 째 부재중 전화.", "scene text: 갯바위에 앉아있는 수리영." ], "visual_inferences": [ "fixed_element character_state: C01 has a slightly exhausted demeanor and subtly damp hair", "fixed_element environment_state: cool desaturated dusk ambience and thick, dense fog", "P01 illuminated screen creates a bright contrast against the dim fog" ], "creative_decisions": [ "Keep the background heavily diffused to emphasize the phone screen and gripping hand", "Frame cuts off most of C01's body to comply with close framing scale while showing a portion of her torso for context" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "f9fc43d30caf5c5a", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "33705be6a112bceb", "owned_usage_hash": "4f53cda18c2baa0c", "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", "P01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 26, "shot_index": 2 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Hold at 수리영's hand height beside her, angled squarely enough for the phone display to remain legible while the device occupies less than two-fifths of the close frame. Her gripping fingers and part of her seated body provide scale around the screen showing 혜수's name and repeated missed-call notifications.", "lighting_mood": "Cool desaturated dusk ambience and dense fog keep the background subdued while the readable phone display provides the shot's local brightness.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": null, "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", "P01" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [ { "subject_id": "C01", "policy_type": "identity_reference", "policy": "base_id_required", "reason": "Only 수리영's hand and a limited portion of her body are visible, so a full outlook reference would not align reliably." } ], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "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", "describe the close subject and its immediate surroundings without relying on the absent background reference", "use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "foggy_dusk_harbor_environment", "element_type": "environment_state", "character_name": "", "description": "The harbor is enveloped in thick, dense fog during the late afternoon or dusk. The ambient lighting is cool and desaturated, creating a subdued and moody atmosphere that significantly reduces background visibility.", "applies_to_shots": [ 2, 5 ], "element_scope": "full", "source_facts": [ "비는 그치고 안개가 자욱하다", "Cool desaturated dusk ambience and dense fog", "Cool desaturated dusk light softened by the harbor fog" ], "visual_inferences": [ "The dense fog completely obscures distant boats or structures, leaving only a soft, gray background.", "The overall color palette is dominated by muted, cool tones typical of dusk after rain." ], "creative_decisions": [ "Keep the background out of focus and heavily diffused by the fog to tightly frame and emphasize the subjects in the foreground." ], "confidence": "high" }, { "element_id": "suri_young_base_appearance", "element_type": "character_state", "character_name": "수리영", "description": "A young Korean woman displaying a slightly exhausted demeanor, with her hair subtly damp from the preceding rain and heavy ambient fog. She wears casual, practical clothing suitable for a chilly, damp evening at the harbor.", "applies_to_shots": [ 2, 5 ], "element_scope": "full", "source_facts": [ "소나기가 한차례 퍼 붓는가 싶더니", "포구를 몇 시간째 돌아다니다 지친", "part of her seated body provide scale (Shot 2)" ], "visual_inferences": [ "Her clothing and hair show subtle signs of moisture due to the rain and thick fog mentioned in the scene.", "Though only parts of her body and face are visible across the different close-up shots, her overall clothing texture and color remain consistent." ], "creative_decisions": [ "Use muted colors for her attire to seamlessly blend with the cool, desaturated aesthetic of the foggy 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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character", "id": "C01", "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" ] }, "render_contracts": [], "_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": "57224c50d32eedb6" }, { "scene_index": 26, "heading": "S#26. 포구 / 해질 무렵", "beat_title": "이동 조력 관계의 성립과 수리영의 간청", "representative_moment": "안개가 자욱한 포구에서 밧줄을 쥔 남자의 팔뚝을 곁에 두고, 그를 바짝 올려다보는 C01의 결연한 얼굴.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Tight framing from chest height, slightly high-angle, cool desaturated dusk lighting", "t2i_prompt": "Photorealistic cinematic still. Seen from chest height with a slight high-angle perspective. C01O01 in a casual hooded jumper, a young Korean woman with an oval face and tied black hair subtly damp from the fog, occupies the center-left frame. Her neck and upturned face are fully visible at three-quarter angle from the right as she leans forward, looking up with a determined expression, her eyes fixed on the man off-screen. At the right frame edge, the forearm and rope-gripping hand of a young Asian man in his 20s enter the foreground, firmly holding a thick mooring rope. [: soft gray blur of thick dense harbor fog at dusk, cool desaturated ambient light, isolating the face with restrained low-key contrast and obscuring all distant elements].", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "밧줄을 쥐고 선 인우를 바짝 올려다보는 수리영의 결연한 얼굴", "비는 그치고 안개가 자욱하다", "인우(24세. 남)는 닷줄을 고정시키고 있다" ], "visual_inferences": [ "Suri-young's hair is subtly damp from the dense fog, carrying the condition from prior shots", "Due to the tight framing and camera height, only In-woo's hand and forearm are visible as foreground context" ], "creative_decisions": [ "Used a tight frame isolating Suri-young's face to emphasize her determination and the claustrophobic feeling of the fog", "Applied a cool desaturated dusk palette with restrained low-key contrast to match the subdued mood" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "f2f0ea48fdfca94b", "owned_hash": "e3b0c44298fc1c14", "camera_direction_hash": "5cd55a378fba1700", "owned_usage_hash": "4f53cda18c2baa0c", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 26, "dependency_reason": "Shot 2의 동일한 안개 낀 포구 배경 유지 (단, 본 샷에서는 close framing으로 인해 배경이 blur 처리됨)", "_shot_index": 5, "visible_entities": [ "C01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 26, "shot_index": 5 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From chest height close to 인우's side, finish the dolly on 수리영's upturned face in a tight, slightly high-angle composition. Her face occupies most of the center-left while only 인우's rope-gripping hand and forearm remain at the right edge as immediate spatial context.", "lighting_mood": "Cool desaturated dusk light softened by the harbor fog isolates 수리영's determined expression with restrained low-key contrast.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": null, "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" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "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", "describe the close subject and its immediate surroundings without relying on the absent background reference", "use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "foggy_dusk_harbor_environment", "element_type": "environment_state", "character_name": "", "description": "The harbor is enveloped in thick, dense fog during the late afternoon or dusk. The ambient lighting is cool and desaturated, creating a subdued and moody atmosphere that significantly reduces background visibility.", "applies_to_shots": [ 2, 5 ], "element_scope": "full", "source_facts": [ "비는 그치고 안개가 자욱하다", "Cool desaturated dusk ambience and dense fog", "Cool desaturated dusk light softened by the harbor fog" ], "visual_inferences": [ "The dense fog completely obscures distant boats or structures, leaving only a soft, gray background.", "The overall color palette is dominated by muted, cool tones typical of dusk after rain." ], "creative_decisions": [ "Keep the background out of focus and heavily diffused by the fog to tightly frame and emphasize the subjects in the foreground." ], "confidence": "high" }, { "element_id": "suri_young_base_appearance", "element_type": "character_state", "character_name": "수리영", "description": "A young Korean woman displaying a slightly exhausted demeanor, with her hair subtly damp from the preceding rain and heavy ambient fog. She wears casual, practical clothing suitable for a chilly, damp evening at the harbor.", "applies_to_shots": [ 2, 5 ], "element_scope": "full", "source_facts": [ "소나기가 한차례 퍼 붓는가 싶더니", "포구를 몇 시간째 돌아다니다 지친", "part of her seated body provide scale (Shot 2)" ], "visual_inferences": [ "Her clothing and hair show subtle signs of moisture due to the rain and thick fog mentioned in the scene.", "Though only parts of her body and face are visible across the different close-up shots, her overall clothing texture and color remain consistent." ], "creative_decisions": [ "Use muted colors for her attire to seamlessly blend with the cool, desaturated aesthetic of the foggy environment." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 26, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "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" } ], "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" ] }, "render_contracts": [], "_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": "42baa1fb1d16ffeb" }, { "scene_index": 27, "heading": "S#27. 옥탑방 안 / 해질 무렵", "beat_title": "수리영의 녹음메시지를 확인하며 굳어버린 혜수", "representative_moment": "엉망이 된 방 한가운데서 굳은 얼굴로 P01을 귀에 댄 채 시선을 떨구고 있는 C02.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Eye-level medium framing focused on Hyesu's upper body, capturing her trapped stillness among the disarray with cool desaturated dusk ambience.", "t2i_prompt": "Photorealistic cinematic still. C02O02 in a casual leather jacket, a Korean middle-aged woman with dark hair, occupies the center from the waist up at three-quarter angle from the left, her gaze directed downward with a rigid expression. Her hand presses P01 against her ear, the smartphone's dark exterior back turned toward the camera. The dark room is bathed in cool desaturated dusk ambience, with restrained low-key contrast falling across her face. A fallen chair and scattered household goods rest on the floor along the lower frame edges, forming low diagonal shapes that surround her still figure.", "outfit_assignments": [ { "character_id": "C02", "outlook_id": "O02" } ], "source_facts": [ "S#27 action: 옥탑방 안에서 수리영의 녹음메시지를 확인하는 혜수.", "shot description: 가재도구가 엉망으로 널브러진 어두운 옥탑방 거실, 휴대전화를 귀에 댄 채 미간을 찌푸린 혜수의 굳은 얼굴." ], "visual_inferences": [ "C02 is positioned in a three-quarter left angle, looking down.", "P01 is pressed against her ear with the screen facing away from the camera.", "The surrounding environment is lit by a dim dusk ambient light." ], "creative_decisions": [ "Framed the shot from the waist up to clearly show her rigid posture anchored amidst the chaotic floor.", "Applied cool desaturated tones and low-key contrast to emphasize the suspenseful and eerie atmosphere of the room's aftermath." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [], "reference_phrase_kinds": [ "character", "background" ], "owned_object_usage": [ { "owned_token": "cabinet", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "chair", "usage_kind": "anchor", "source_phrase": "A fallen chair and scattered household goods rest on the floor along the lower frame edges" }, { "owned_token": "curtain", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "dining table", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "drawer", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "entrance door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "floor lamp", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "footprint", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "household goods", "usage_kind": "anchor", "source_phrase": "A fallen chair and scattered household goods rest on the floor along the lower frame edges" }, { "owned_token": "living-room window", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "sofa", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "4e3182e30a4a2a37", "owned_hash": "108e3b65bc2770f6", "camera_direction_hash": "81e339a94ab7fc6a", "owned_usage_hash": "3c2063e2364c0df2", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "chair", "violating_phrase": "A fallen chair and scattered household goods rest on the floor along the lower frame edges", "reason": "The prompt mentions the fallen chair to establish composition and framing at the frame boundaries, which correctly aligns with the producer's anchor declaration.", "verdict": "anchor_reference" }, { "owned_object": "household goods", "violating_phrase": "A fallen chair and scattered household goods rest on the floor along the lower frame edges", "reason": "The scattered household goods are referenced purely to guide the composition and framing in the lower frame, matching the producer's anchor declaration.", "verdict": "anchor_reference" } ] } } ], "scene_type": "normal", "dependent_scene_index": 14, "dependency_reason": "동일한 옥탑방 거실 공간이므로 이전 씬(S14_Shot3)의 배경 구조와 벽/가구 레이아웃을 일관되게 유지하기 위함", "_shot_index": 1, "visible_entities": [ "L05", "C02", "P01" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 27, "shot_index": 1 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "medium", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Complete the eye-level approach but settle slightly wider than a conventional facial close-up, keeping 혜수's rigid upper body centered among the room's disarray. Her phone stays pressed to her ear while fallen chairs and disturbed household items remain visible along both lower edges, making her stillness feel trapped within the aftermath.", "lighting_mood": "The dark rooftop room is held in cool desaturated dusk ambience with restrained low-key contrast around 혜수's rigid face.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": null, "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": [ "L05", "C02", "P01" ], "allowed_outlook_pairs": [ { "character_id": "C02", "outlook_id": "O02" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L05B04", "reference_usage": "exact_background", "owned_objects": [ "cabinet", "chair", "curtain", "dining table", "drawer", "entrance door", "floor lamp", "footprint", "household goods", "living-room window", "sofa" ], "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [], "previous_shot_refs": [ { "scene_index": 14, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C02O02", "policy": "required" }, { "kind": "background", "id": "L05B04", "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" ] }, "render_contracts": [], "_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": "324e331c31b6c1aa" }, { "scene_index": 28, "heading": "S#28. 바다 + 갑판 / 해질 무렵", "beat_title": "바다 한가운데로 향하는 어선", "representative_moment": "해질 무렵 붉은 노을이 깔린 바다 위를 전진하는 낡은 소형 어선의 원경.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Static low-angle wide shot just above water level on the forward axis", "t2i_prompt": "Photorealistic cinematic still. A small fishing boat occupies the middle-center midground, moving toward the camera along the bow axis. Observed from just above the water level, the approaching vessel appears small against the vast negative space of the sea and sky. The hull's full exterior silhouette is preserved against the horizon, with symmetrical white spray opening from either side of the bow in a forward-driving wedge. The wheelhouse and mast stand steady atop the deck. Cool desaturated sea tones anchor the lower frame, contrasting with the deep-red and muted-gold light of the sunset filling the upper sky, maintaining restrained dramatic contrast.", "outfit_assignments": [], "source_facts": [ "shot description: Wide exterior shot of the small fishing boat advancing on the sea at dusk, symmetrical white spray on both sides of the bow.", "scene text: 바다위에 떠있는 소형어선. 한참 떠나온 듯 하다." ], "visual_inferences": [ "fixed_element dusk_lighting_atmosphere: The environment is immersed in a natural dusk setting with consistent ambient lighting.", "The cool desaturated sea tones contrast with the deep-red and muted-gold sunset." ], "creative_decisions": [ "Placed the observer just above water level to emphasize the smallness of the boat against the vast sea, aligning with the DP's forward-axis omniscient perspective.", "Centered the approaching bow to highlight the symmetrical white spray and forward-driving wedge motion." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001" ], "reference_phrase_kinds": [ "background" ], "owned_object_usage": [ { "owned_token": "antenna", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "buoy", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "capstan", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "exhaust pipe", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "fish box", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "fishing boat", "usage_kind": "anchor", "source_phrase": "A small fishing boat occupies the middle-center midground" }, { "owned_token": "hull", "usage_kind": "anchor", "source_phrase": "The hull's full exterior silhouette is preserved against the horizon" }, { "owned_token": "mast", "usage_kind": "anchor", "source_phrase": "The wheelhouse and mast stand steady atop the deck" }, { "owned_token": "navigation light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "net", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "railing", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "rope", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "searchlight", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "tire fender", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wheelhouse", "usage_kind": "anchor", "source_phrase": "The wheelhouse and mast stand steady atop the deck" }, { "owned_token": "winch", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "ccec08db1344a299", "owned_hash": "064ef7f357eb3efa", "camera_direction_hash": "faf341d3a7a69eac", "owned_usage_hash": "ae5072e04259427f", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "fishing boat", "violating_phrase": "A small fishing boat occupies the middle-center midground", "reason": "The fishing boat is mentioned to establish its position and movement in the composition, which acts as a spatial anchor and matches the producer's anchor declaration.", "verdict": "anchor_reference" }, { "owned_object": "hull", "violating_phrase": "The hull's full exterior silhouette is preserved against the horizon", "reason": "The hull is referenced to describe its silhouette positioning relative to the horizon, which functions as a composition anchor and matches the producer's anchor declaration.", "verdict": "anchor_reference" }, { "owned_object": "wheelhouse", "violating_phrase": "The wheelhouse and mast stand steady atop the deck", "reason": "The wheelhouse is described as standing steady to establish compositional stability, matching the producer's declaration as an anchor.", "verdict": "anchor_reference" }, { "owned_object": "mast", "violating_phrase": "The wheelhouse and mast stand steady atop the deck", "reason": "The mast is referenced alongside the wheelhouse to anchor the visual composition on the deck, matching the producer's anchor declaration.", "verdict": "anchor_reference" } ] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "", "_shot_index": 1, "visible_entities": [ "L18", "L19" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 28, "shot_index": 1 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "wide", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Place the static camera at a long distance just above water level on the vessel's forward axis, with the approaching bow centered and the boat kept small against the breadth of sea and sunset sky. Symmetrical white spray opens from either side of the bow, creating a forward-driving wedge while preserving the full exterior silhouette.", "lighting_mood": "Cool desaturated sea tones contrast with the scene-grounded deep-red and muted-gold sunset, maintaining restrained dramatic contrast.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "movement_direction", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "midground", "gesture_action": "moves_toward", "gesture_target_label": "camera along the bow axis", "label": "small fishing boat", "screen_zone": "middle_center", "target_id": "", "target_kind": "background" } ] }, "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": [ "L18", "L19" ], "allowed_outlook_pairs": [], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L19B03", "reference_usage": "exact_background", "owned_objects": [ "antenna", "buoy", "capstan", "door", "exhaust pipe", "fish box", "fishing boat", "hull", "mast", "navigation light", "net", "railing", "rope", "searchlight", "tire fender", "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "dusk_lighting_atmosphere", "element_type": "environment_state", "character_name": "", "description": "The environment is immersed in a natural dusk setting with consistent ambient lighting. The overall color palette remains cool, desaturated, and low-key across both exterior and interior spaces. This bluish atmospheric tone subtly contrasts with the deep-red and muted-gold colors of the sunset.", "applies_to_shots": [ 1, 5, 10 ], "element_scope": "full", "source_facts": [ "Cool desaturated sea tones contrast with the scene-grounded deep-red and muted-gold sunset (Shot 1)", "Natural dusk ambience is kept cool and desaturated (Shot 5)", "Natural dusk ambience inside the wheelhouse remains cool, desaturated, and low-key (Shot 10)" ], "visual_inferences": [ "The dim, cool ambient light uniformly affects the exterior of the boat, the interior of the wheelhouse, and the character's visible skin tones." ], "creative_decisions": [ "Maintaining restrained contrast and a desaturated cool palette to emphasize a tense and desolate mood throughout the sequence." ], "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "background", "id": "L19B03", "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" ] }, "render_contracts": [], "_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": "47139d1269c91f3d" }, { "scene_index": 28, "heading": "S#28. 바다 + 갑판 / 해질 무렵", "beat_title": "사진을 보고 공포에 질린 인우", "representative_moment": "사진을 본 후 눈을 크게 뜨고 당황하며 굳어버린 C04의 얼굴 클로즈업.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "tight face close-up at eye level, cool desaturated dusk lighting with restrained contrast", "t2i_prompt": "Photorealistic cinematic still. C04O04 in a windbreaker, a Korean young man, captured in a tight face close-up at three-quarter angle from the right. His head recoils slightly, looking down with widened pupils fixed on the lower edge. A narrow, softly focused edge of a photograph enters the bottom of the frame, its image-bearing surface angled upward toward him. The scene is illuminated by cool, desaturated natural dusk ambient light with restrained contrast concentrating attention on his eyes, while the background dissolves into a soft blur carrying faint deep-red and muted-gold tones.", "outfit_assignments": [ { "character_id": "C04", "outlook_id": "O04" } ], "source_facts": [ "S#28 text: 인우 (사진보더니 당황)", "Shot description: 사진을 보고 동공이 크게 확장된 채 굳어있는 인우의 당황한 얼굴 클로즈업" ], "visual_inferences": [ "fixed_element dusk_lighting_atmosphere: Cool, desaturated dusk ambience with deep-red and muted-gold tones", "Photograph remains softly focused at the bottom of the frame to keep focus on the character's reaction" ], "creative_decisions": [ "Tight framing on the face to emphasize the widened pupils and recoil", "Background reduced to an out-of-focus blur to adhere to the skipped_close_framing background rule and isolate the subject" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [ { "owned_token": "antenna", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "bollard", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "capstan", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "deck", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "drain", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "fishing net", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "lifebuoy", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "navigation light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "pipe", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "railing", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "rope", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "storage box", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wheelhouse", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "winch", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "work light", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "8fca2cb33b03d954", "owned_hash": "8221ad03addc79b1", "camera_direction_hash": "112eb0b9ae46225e", "owned_usage_hash": "f36b06ff0e07a93f", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 28, "dependency_reason": "Shot 1 established the natural dusk setting and overall cool, desaturated sea tones contrasting with the sunset.", "_shot_index": 5, "visible_entities": [ "C04" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 28, "shot_index": 5 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "End the push at 인우's eye level in a tight face close-up, favoring a slight three-quarter angle rather than a flat frontal portrait. His widened pupils hold near the center while a narrow, softly focused edge of the presented photograph remains low in frame to preserve the cause of his alarm.", "lighting_mood": "Natural dusk ambience is kept cool and desaturated, with restrained contrast concentrating attention on 인우's widened eyes.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": null, "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": [ "C04" ], "allowed_outlook_pairs": [ { "character_id": "C04", "outlook_id": "O04" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "L19B02", "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", "describe the close subject and its immediate surroundings without relying on the absent background reference", "use [L##: ...] block for frame-edge surface description" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "dusk_lighting_atmosphere", "element_type": "environment_state", "character_name": "", "description": "The environment is immersed in a natural dusk setting with consistent ambient lighting. The overall color palette remains cool, desaturated, and low-key across both exterior and interior spaces. This bluish atmospheric tone subtly contrasts with the deep-red and muted-gold colors of the sunset.", "applies_to_shots": [ 1, 5, 10 ], "element_scope": "full", "source_facts": [ "Cool desaturated sea tones contrast with the scene-grounded deep-red and muted-gold sunset (Shot 1)", "Natural dusk ambience is kept cool and desaturated (Shot 5)", "Natural dusk ambience inside the wheelhouse remains cool, desaturated, and low-key (Shot 10)" ], "visual_inferences": [ "The dim, cool ambient light uniformly affects the exterior of the boat, the interior of the wheelhouse, and the character's visible skin tones." ], "creative_decisions": [ "Maintaining restrained contrast and a desaturated cool palette to emphasize a tense and desolate mood throughout the sequence." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 28, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C04O04", "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" ] }, "render_contracts": [], "_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": "da252caa2f2c1785" }, { "scene_index": 28, "heading": "S#28. 바다 + 갑판 / 해질 무렵", "beat_title": "조타기를 향해 뻗은 인우의 팔을 수리영이 꽉 움켜쥠", "representative_moment": "수리영(C01)이 조타기를 돌리려는 인우(C04)의 팔뚝을 양손으로 꽉 움켜쥔 순간", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Insert framing at forearm height, capturing the intense tactile struggle under cool, desaturated dusk lighting", "t2i_prompt": "Photorealistic cinematic still. Captured at forearm height, C01 in a dark hooded jacket, a young Korean woman, occupies the middle-left foreground reaching for the forearm of C04 in a fisher windbreaker, a young Korean man, who occupies the middle-right foreground reaching toward the helm in the upper-right background. C01 is framed partially from over the shoulder with her gaze directed down, bracing both hands forward to form a tight, compressing grip on C04's forearm, the locked grip crossing the lower-center of the frame. C04 is partially visible looking away off-screen, the fabric of his sleeve visibly indented by the intense tactile pressure. The operating side of the helm faces C04 as the interrupted destination of his reach. The interior wheelhouse surfaces recede behind the interlocked arms, anchoring them in the same shared space. The natural dusk ambience remains cool, desaturated, and low-key, casting a dim, bluish atmospheric tone uniformly across the visible skin and interior.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C04", "outlook_id": "O04" } ], "source_facts": [ "Shot 10 description: 인우 reaches for the helm, and 수리영's two hands tightly grip his reaching forearm.", "수리영 stops 인우 inside the wheelhouse.", "인우 tries to look away, evading the reality." ], "visual_inferences": [ "fixed_element dusk_lighting_atmosphere: The environment is immersed in a natural dusk setting with consistent ambient lighting.", "The tight grip creates a visible physical indentation on the clothing/skin, emphasizing the unequal tension.", "The camera angle isolates the forearms and hands, keeping the rest of the bodies partially out of frame to match the insert framing." ], "creative_decisions": [ "Positioning the interlocked arms in the lower-center while keeping the helm in the upper-right background creates a clear spatial line of interrupted action.", "Using a cool, desaturated, and low-key palette to emphasize the suspenseful and bleak mood." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002", "fsc_003" ], "reference_phrase_kinds": [ "character", "background" ], "owned_object_usage": [ { "owned_token": "cable", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "ceiling light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "deck", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "helm", "usage_kind": "anchor", "source_phrase": "reaching toward the helm in the upper-right background" }, { "owned_token": "instrument panel", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "pipe", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "radio", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "railing", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "steering wheel", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "throttle", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "windshield wiper", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "a8ecdc8242a54400", "owned_hash": "32d8cfe4aef76fe9", "camera_direction_hash": "18d86750e9bc4a38", "owned_usage_hash": "d83713262fd3add7", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "helm", "violating_phrase": "reaching toward the helm in the upper-right background", "reason": "The helm is used purely as a spatial destination and compositional anchor for C04's reach, which matches the producer's anchor declaration.", "verdict": "anchor_reference" } ] } } ], "scene_type": "normal", "dependent_scene_index": 28, "dependency_reason": "Same physical location (wheelhouse interior) from previous shots in Scene 28, continuing the dusk lighting atmosphere.", "_shot_index": 10, "visible_entities": [ "C01", "C04" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 28, "shot_index": 10 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "insert", "moment_lock": "single still moment from the selected shot only", "camera_direction": "Lower to forearm height beside the helm and finish the dolly on the point of contact, isolating 인우's reaching forearm and 수리영's two compressing hands as an insert. The helm remains in the upper-right background as the physical destination of his reach, while the locked grip crosses the lower center without enlarging any nearby object unnaturally.", "lighting_mood": "Natural dusk ambience inside the wheelhouse remains cool, desaturated, and low-key, emphasizing the tactile pressure of hands against forearm.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "shared_space_relation", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "background", "gesture_action": "none", "gesture_target_label": "", "label": "helm", "screen_zone": "upper_right", "target_id": "", "target_kind": "background" }, { "constraint_id": "fsc_002", "depth_plane": "foreground", "gesture_action": "reaches_for", "gesture_target_label": "인우's forearm", "label": "수리영", "screen_zone": "middle_left", "target_id": "C01", "target_kind": "character" }, { "constraint_id": "fsc_003", "depth_plane": "foreground", "gesture_action": "reaches_for", "gesture_target_label": "helm", "label": "인우", "screen_zone": "middle_right", "target_id": "C04", "target_kind": "character" } ] }, "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" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C04", "outlook_id": "O04" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [ { "subject_id": "C01", "policy_type": "identity_reference", "policy": "base_id_required", "reason": "Only 수리영's gripping hands and partial arm orientation are visible, making a full outlook reference unsuitable." }, { "subject_id": "C04", "policy_type": "identity_reference", "policy": "base_id_required", "reason": "Only 인우's forearm is visible, so identity should rely on the base reference rather than a full outlook." } ], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L19B01", "reference_usage": "exact_background", "owned_objects": [ "cable", "ceiling light", "deck", "door", "helm", "instrument panel", "pipe", "radio", "railing", "steering wheel", "throttle", "window", "windshield wiper" ], "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "dusk_lighting_atmosphere", "element_type": "environment_state", "character_name": "", "description": "The environment is immersed in a natural dusk setting with consistent ambient lighting. The overall color palette remains cool, desaturated, and low-key across both exterior and interior spaces. This bluish atmospheric tone subtly contrasts with the deep-red and muted-gold colors of the sunset.", "applies_to_shots": [ 1, 5, 10 ], "element_scope": "full", "source_facts": [ "Cool desaturated sea tones contrast with the scene-grounded deep-red and muted-gold sunset (Shot 1)", "Natural dusk ambience is kept cool and desaturated (Shot 5)", "Natural dusk ambience inside the wheelhouse remains cool, desaturated, and low-key (Shot 10)" ], "visual_inferences": [ "The dim, cool ambient light uniformly affects the exterior of the boat, the interior of the wheelhouse, and the character's visible skin tones." ], "creative_decisions": [ "Maintaining restrained contrast and a desaturated cool palette to emphasize a tense and desolate mood throughout the sequence." ], "confidence": "high" } ], "previous_shot_refs": [ { "scene_index": 28, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character", "id": "C01", "policy": "required" }, { "kind": "character", "id": "C04", "policy": "required" }, { "kind": "background", "id": "L19B01", "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" ] }, "render_contracts": [], "_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": "435ec6ae6f2a1e0b" }, { "scene_index": 29, "heading": "S#29. 조타실 / 해질 무렵", "beat_title": "사진 속 엄마의 이름", "representative_moment": "수리영의 손가락이 피 묻은 낡은 사진(P03) 속 어린 소녀의 형상을 꾹 누르고 있다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Overhead framing, extremely close to the surface, cool and heavily desaturated dusk lighting with deep-crimson accents", "t2i_prompt": "Photorealistic cinematic still. Overhead framing from directly above at hand height, extremely close to the surface. C01 in a casual hooded jacket, an Asian 21-year-old woman, stands implied beyond the frame edges with her gaze directed off-screen, her hand and pressing fingertip sharply focused in the center of the composition. The fingertip presses firmly against P03, an old, faded vintage photographic-paper print occupying less than the lower half of the frame. The image-bearing front face of the photograph reveals the reproduced figure of a 12-year-old Korean girl positioned directly beneath the pressing finger, with deep-crimson blood marks dotting the paper around it. The photograph rests flat on the dark instrument panel of the console. Restrained ambient illumination for dusk casts a cool, heavily desaturated color palette across the space, keeping the low-key contrast sharp while the blood marks serve as a selective deep-crimson accent. The edges of the hand provide the only depth cue against the subdued background.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" } ], "source_facts": [ "S#29 action: 수리영 points at the photo of her mother.", "Shot description: 수리영's hand and fingertip press firmly against the girl's image on the old photograph.", "DP instruction: The camera is directly above the photograph at hand height." ], "visual_inferences": [ "fixed_element wheelhouse_dusk_lighting: The space maintains a cool, heavily desaturated color palette suitable for dusk.", "Only the hand is framed, meaning C01's body and off-screen gaze are implied outside the frame limits.", "The photograph rests on the instrument panel of the console." ], "creative_decisions": [ "Used an extremely close overhead angle to emphasize the physical contact between the fingertip and the reproduced figure without explicitly stating exact distances.", "Highlighted the deep-crimson blood marks against the cool desaturated background to amplify the ominous tone." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [], "reference_phrase_kinds": [ "character", "prop", "background" ], "owned_object_usage": [ { "owned_token": "ceiling light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "console", "usage_kind": "anchor", "source_phrase": "console" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "helm", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "instrument panel", "usage_kind": "anchor", "source_phrase": "dark instrument panel" }, { "owned_token": "photograph", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "radio", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "red circle marking", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "switchboard", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wall", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "2ec50a4959baffcc", "owned_hash": "a1fe2114dd4fa7c1", "camera_direction_hash": "05790346ccd3fd5f", "owned_usage_hash": "62c3f5528c28f6f0", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "console", "violating_phrase": "on the dark instrument panel of the console", "reason": "The console is referenced purely as a physical placement anchor for the photograph, which aligns with the producer's anchor declaration.", "verdict": "anchor_reference" }, { "owned_object": "instrument panel", "violating_phrase": "dark instrument panel", "reason": "The instrument panel is used solely as a spatial support surface for the photograph, matching the producer's anchor declaration.", "verdict": "anchor_reference" }, { "owned_object": "photograph", "violating_phrase": "The image-bearing front face of the photograph reveals the reproduced figure of a 12-year-old Korean girl... with deep-crimson blood marks dotting the paper around it", "reason": "The visual content within the photograph is described in detail as required by the shot_intent ('피 묻은 낡은 사진(P03) 속 어린 소녀의 형상'), satisfying exception (1) for allowed visual state change, despite the producer's incorrect declaration of it being absent.", "verdict": "allowed_visual_state_change" } ] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "No dependent scene indicated for this specific insert shot.", "_shot_index": 3, "visible_entities": [ "C01", "P03" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 29, "shot_index": 3 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "insert", "moment_lock": "single still moment from the selected shot only", "camera_direction": "At the end of the dolly-in, the camera is directly above the photograph at hand height, roughly 20 cm away, with 수리영's pressing fingertip and the photographed girl's figure held in the center while the old photograph occupies less than the lower half of frame. The near-overhead insert keeps the blood marks and finger pressure sharp, with the edges of 수리영's hand providing the only depth cue before the camera is ready to tilt upward.", "lighting_mood": "Restrained ambient illumination appropriate to the wheelhouse at dusk keeps the palette cool and desaturated while the blood remains a selective deep-crimson accent.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": null, "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", "P03" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [ { "subject_id": "C01", "policy_type": "identity_reference", "policy": "base_id_required", "reason": "Only 수리영's hand and fingertip are framed, so a full outlook reference would not align with the visible anatomy." } ], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L20B02", "reference_usage": "exact_background", "owned_objects": [ "ceiling light", "chair", "console", "door", "helm", "instrument panel", "photograph", "radio", "red circle marking", "switchboard", "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "continuity_elements_used": { "fixed_elements": [ { "element_id": "wheelhouse_dusk_lighting", "element_type": "environment_state", "character_name": "", "description": "The interior environment is bathed in restrained, low ambient lighting suitable for dusk. This illumination creates a cool, heavily desaturated color palette across the space, maintaining a subdued atmosphere with low-key contrast.", "applies_to_shots": [ 3, 6, 9 ], "element_scope": "full", "source_facts": [ "Restrained ambient illumination appropriate to the wheelhouse at dusk keeps the palette cool and desaturated (Shot 3)", "Low ambient illumination appropriate to the wheelhouse at dusk preserves restrained contrast (Shot 6)", "Restrained dusk-appropriate ambient light renders a cool desaturated field with low-key contrast (Shot 9)" ], "visual_inferences": [], "creative_decisions": [ "Maintained a consistent cool and desaturated lighting tone to visually connect the disparate close-up subjects into the same spatial reality." ], "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character", "id": "C01", "policy": "required" }, { "kind": "background", "id": "L20B02", "policy": "required" }, { "kind": "prop", "id": "P03", "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" ] }, "render_contracts": [ { "contract_id": "rc_001", "scope": { "scene_index": 29, "shot_indices": [ 3 ], "duration": "single_shot" }, "targets": [ { "entity_id": "P03", "role": "visual_target" } ], "requirements": [ { "dimension": "visual_identity", "operation": "preserve", "strength": "required", "reference_policy": "use_entity_reference" } ] } ], "_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": "bfccab0d592300ef" }, { "scene_index": 29, "heading": "S#29. 조타실 / 해질 무렵", "beat_title": "기이한 붉은 원의 발견과 수리영의 실신", "representative_moment": "조타실 벽면에 빼곡하게 그려진 기이한 붉은 원 표식들이 비스듬한 시선으로 포착된다.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "oblique close-up focusing on receding wall textures with cool desaturated dusk lighting", "t2i_prompt": "Photorealistic cinematic still. Close framing from an oblique observer viewpoint at eye level, positioned close to the surface. The composition is dominated by a frame-edge surface: [L20: an old wooden interior wall of the wheelhouse], presented obliquely to the camera. The wooden surface is densely covered with numerous strange P02, their circular shapes receding toward one edge of the frame in an oppressive repetition rather than a flat pattern. The interior environment is bathed in restrained, low ambient lighting suitable for dusk, creating a cool, heavily desaturated color palette across the space. This illumination maintains a subdued atmosphere with low-key contrast, leaving the wooden wall shadowed while isolating the P02 as stark deep-crimson accents.", "outfit_assignments": [], "source_facts": [ "shot description: Close-up of bizarre red circular shapes densely drawn on the old wooden wall of the wheelhouse", "DP instruction: The track ends just behind and beside the position formerly occupied by 인우's shoulder... body now cleared from the composition", "Low ambient illumination appropriate to the wheelhouse at dusk keeps the palette cool and desaturated" ], "visual_inferences": [ "fixed_element wheelhouse_dusk_lighting: The interior environment is bathed in restrained, low ambient lighting suitable for dusk, creating a cool, heavily desaturated color palette", "The oblique angle turns the repeated circular markings into a receding, threatening pattern rather than a flat decorative design" ], "creative_decisions": [ "Applied close framing to skip the wider background reference, focusing purely on the claustrophobic accumulation of red marks on the frame-edge wall surface", "Emphasized the deep-crimson tone of P02 against the cool, low-key contrast of the background to enhance the uncanny horror atmosphere" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [], "reference_phrase_kinds": [], "owned_object_usage": [ { "owned_token": "ceiling light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "console", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "helm", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "instrument panel", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "photograph", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "radio", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "red circle marking", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "switchboard", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wall", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "86f03b0543e65834", "owned_hash": "a1fe2114dd4fa7c1", "camera_direction_hash": "0536c52e068c4cd6", "owned_usage_hash": "d965da3af9825c89", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "", "_shot_index": 6, "visible_entities": [ "L20", "P02" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 29, "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 track ends just behind and beside the position formerly occupied by 인우's shoulder, at eye level and about 50 cm from the wall, with his body now cleared from the composition. A slightly oblique close view lets successive red circles recede toward one edge rather than flattening them into a decorative frontal pattern, holding on their oppressive repetition before the pan away.", "lighting_mood": "Low ambient illumination appropriate to the wheelhouse at dusk preserves restrained contrast, with the wall subdued and the circles isolated as deep-crimson accents.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": null, "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": [ "L20", "P02" ], "allowed_outlook_pairs": [], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "L20B02", "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", "describe the close subject and its immediate surroundings without relying on the absent background reference", "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 interior environment is bathed in restrained, low ambient lighting suitable for dusk. This illumination creates a cool, heavily desaturated color palette across the space, maintaining a subdued atmosphere with low-key contrast.", "applies_to_shots": [ 3, 6, 9 ], "element_scope": "full", "source_facts": [ "Restrained ambient illumination appropriate to the wheelhouse at dusk keeps the palette cool and desaturated (Shot 3)", "Low ambient illumination appropriate to the wheelhouse at dusk preserves restrained contrast (Shot 6)", "Restrained dusk-appropriate ambient light renders a cool desaturated field with low-key contrast (Shot 9)" ], "visual_inferences": [], "creative_decisions": [ "Maintained a consistent cool and desaturated lighting tone to visually connect the disparate close-up subjects into the same spatial reality." ], "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [], "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" ] }, "render_contracts": [], "_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": "d33ec791d3ee19b2" }, { "scene_index": 29, "heading": "조타실 / 해질 무렵", "beat_title": "의식을 잃고 쓰러지는 수리영", "representative_moment": "C01 collapses unconscious onto C04's shoulder.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "Close-up handheld framing, shallow depth of field, cool desaturated dusk lighting", "t2i_prompt": "Photorealistic cinematic still. C01O01 in a casual hooded jacket, a 21-year-old Asian woman, occupies the middle-left foreground, while C04O04 in a fishing windbreaker, a 24-year-old Asian man, is positioned in the middle-right midground. Framed at a close distance, C01O01 is captured in sharp left profile, unconscious and limp with her eyes closed, her pale cheek resting heavily against C04O04's shoulder. C04O04 is partially visible on the right edge of the frame in an over-the-shoulder view, his upper body bracing her fallen weight, his eyes fixed on her face. Her long black tied hair spills slightly over the fabric. The space is bathed in restrained dusk-appropriate ambient light, creating a cool, heavily desaturated color palette with low-key facial contrast. The shallow depth of field drops the surroundings away, leaving only [: a soft, out-of-focus wheelhouse wall in subdued twilight] at the frame edge.", "outfit_assignments": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C04", "outlook_id": "O04" } ], "source_facts": [ "Suri-young loses consciousness and collapses onto In-woo's shoulder.", "The scene takes place in a wheelhouse at dusk." ], "visual_inferences": [ "wheelhouse_dusk_lighting element applied to maintain the cool desaturated palette.", "Close framing isolates the physical contact and hides the background details as per skipped_close_framing mode." ], "creative_decisions": [ "Placed C01 on the middle-left and C04 on the middle-right to satisfy the frame spatial contract.", "Used a very shallow depth of field to emphasize C01's face and blur out the background completely." ], "confidence": "high", "applied_frame_spatial_constraint_ids": [ "fsc_001", "fsc_002" ], "reference_phrase_kinds": [ "character" ], "owned_object_usage": [ { "owned_token": "GPS display", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "ceiling light", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "chair", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "compass", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "console", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "door", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "handrail", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "helm", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "instrument panel", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "radar display", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "radio", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "storage cabinet", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "throttle", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "wall panel", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "window", "usage_kind": "absent", "source_phrase": "" }, { "owned_token": "windshield wiper", "usage_kind": "absent", "source_phrase": "" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "573f16b668cb1da7", "owned_hash": "7f5bef5963446ae6", "camera_direction_hash": "c8aa10da2479ca10", "owned_usage_hash": "4e02c3857d33bcbb", "validator": "scene_detail_owned_objects.v1.close_skip", "violations": [] } } ], "scene_type": "normal", "dependent_scene_index": 29, "dependency_reason": "Maintains the same restrained dusk lighting and desaturated color palette established in previous shots of Scene 29.", "_shot_index": 9, "visible_entities": [ "C01", "C04" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 29, "shot_index": 9 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "close", "moment_lock": "single still moment from the selected shot only", "camera_direction": "The handheld camera settles beside 인우's shoulder at 수리영's head height, roughly 25 cm from her face, framing her closed-eyed profile across the left and center while his shoulder and partial head brace the right edge. Her cheek hangs against him in sharp focus as the remaining background falls away, and the final residual sway dies with her body coming to rest.", "lighting_mood": "Restrained dusk-appropriate ambient light renders 수리영's pallor against a cool desaturated field with low-key but readable facial contrast.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": { "reason": "shared_space_relation", "constraints": [ { "constraint_id": "fsc_001", "depth_plane": "foreground", "gesture_action": "none", "gesture_target_label": "", "label": "수리영", "screen_zone": "middle_left", "target_id": "C01", "target_kind": "character" }, { "constraint_id": "fsc_002", "depth_plane": "midground", "gesture_action": "none", "gesture_target_label": "", "label": "인우", "screen_zone": "middle_right", "target_id": "C04", "target_kind": "character" } ] }, "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" ], "allowed_outlook_pairs": [ { "character_id": "C01", "outlook_id": "O01" }, { "character_id": "C04", "outlook_id": "O04" } ], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": false, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "skipped_close_framing", "bg_id": "L20B01", "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", "describe the close subject and its immediate surroundings without relying on the absent background reference", "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 interior environment is bathed in restrained, low ambient lighting suitable for dusk. This illumination creates a cool, heavily desaturated color palette across the space, maintaining a subdued atmosphere with low-key contrast.", "applies_to_shots": [ 3, 6, 9 ], "element_scope": "full", "source_facts": [ "Restrained ambient illumination appropriate to the wheelhouse at dusk keeps the palette cool and desaturated (Shot 3)", "Low ambient illumination appropriate to the wheelhouse at dusk preserves restrained contrast (Shot 6)", "Restrained dusk-appropriate ambient light renders a cool desaturated field with low-key contrast (Shot 9)" ], "visual_inferences": [], "creative_decisions": [ "Maintained a consistent cool and desaturated lighting tone to visually connect the disparate close-up subjects into the same spatial reality." ], "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "character_outlook", "id": "C04O04", "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" ] }, "render_contracts": [], "_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": "2424a250eec6327c" }, { "scene_index": 30, "heading": "S#30. 바다 / 해질 무렵", "beat_title": "안개 속으로 사라지는 배", "representative_moment": "석양이 비치는 바다 위, 짙은 안개 띠의 경계면을 뚫고 들어가는 어선의 앞머리.", "t2i_variations": [ { "variant_label": "var_1", "camera_effect": "wide side view from near water level, cool desaturated palette with restrained golden reflection", "t2i_prompt": "Photorealistic cinematic still. Viewed from near water level off the vessel's side, the fishing boat occupies the middle plane, its bow crossing the sharply legible boundary of the dense fog bank ahead. The wheelhouse, antenna, navigation light, and lifebuoy remain visible on the hull outside the mist. The open sea fills the lower frame, a wake trailing behind the vessel. The water surface reflects the setting sun as a golden trace around the vessel's course across the cool, desaturated sea, while the fog bank remains tonally muted.", "outfit_assignments": [], "source_facts": [ "S#30 action: '서서히 지는 해가 바다위에 황금빛 그림자를 드리운다. 서서히 안개 안으로 들어가는 인우의 배.'", "shot description: 바다 한가운데 자욱하게 깔린 짙은 안개 띠의 경계면을 향해 어선의 앞머리가 뚫고 들어간 순간" ], "visual_inferences": [ "low water-level angle emphasizing the boundary between the open sea and the fog bank", "boat details like the wheelhouse and lifebuoy remain outside the fog to highlight the physical transition into the obscured space" ], "creative_decisions": [ "wide side view from slightly aft of the bow to capture the exact moment of crossing the threshold", "cool desaturated sea contrasted with a muted golden reflection to evoke the claustrophobic suspense of the scene" ], "confidence": "high", "applied_frame_spatial_constraint_ids": [], "reference_phrase_kinds": [ "background" ], "owned_object_usage": [ { "owned_token": "antenna", "usage_kind": "anchor", "source_phrase": "antenna" }, { "owned_token": "fishing boat", "usage_kind": "anchor", "source_phrase": "the fishing boat occupies the middle plane" }, { "owned_token": "fog bank", "usage_kind": "anchor", "source_phrase": "the dense fog bank ahead" }, { "owned_token": "lifebuoy", "usage_kind": "anchor", "source_phrase": "lifebuoy" }, { "owned_token": "navigation light", "usage_kind": "anchor", "source_phrase": "navigation light" }, { "owned_token": "sea", "usage_kind": "anchor", "source_phrase": "The open sea fills the lower frame" }, { "owned_token": "wake", "usage_kind": "anchor", "source_phrase": "a wake trailing behind the vessel" }, { "owned_token": "wheelhouse", "usage_kind": "anchor", "source_phrase": "The wheelhouse" } ], "owned_validation": { "schema_version": 2, "t2i_prompt_hash": "cdb650d39f6a7156", "owned_hash": "8c9cba7920bdfd1a", "camera_direction_hash": "5661e14dea2a401c", "owned_usage_hash": "1d6ca37ef9cdfbee", "validator": "scene_detail_owned_objects.v1", "violations": [ { "owned_object": "antenna", "violating_phrase": "The wheelhouse, antenna, navigation light, and lifebuoy remain visible on the hull outside the mist", "reason": "The prompt mentions the antenna as remaining visible on the hull to anchor its position, matching the producer's 'anchor' classification.", "verdict": "anchor_reference" }, { "owned_object": "fishing boat", "violating_phrase": "the fishing boat occupies the middle plane", "reason": "The prompt specifies the fishing boat's composition and placement in the frame as an anchor, matching the producer's declaration.", "verdict": "anchor_reference" }, { "owned_object": "fog bank", "violating_phrase": "the dense fog bank ahead", "reason": "The prompt anchors the dense fog bank as a spatial boundary, which is consistent with the producer's 'anchor' classification.", "verdict": "anchor_reference" }, { "owned_object": "lifebuoy", "violating_phrase": "lifebuoy remain visible on the hull outside the mist", "reason": "The prompt refers to the lifebuoy remaining visible to establish visibility, matching the producer's 'anchor' classification.", "verdict": "anchor_reference" }, { "owned_object": "navigation light", "violating_phrase": "navigation light, and lifebuoy remain visible on the hull", "reason": "The navigation light is mentioned purely to specify its visibility outside the mist, aligning with the producer's 'anchor' classification.", "verdict": "anchor_reference" }, { "owned_object": "sea", "violating_phrase": "The open sea fills the lower frame", "reason": "The sea is referenced to describe its framing and surface reflections without redrawing it, matching the producer's 'anchor' declaration.", "verdict": "anchor_reference" }, { "owned_object": "wake", "violating_phrase": "a wake trailing behind the vessel", "reason": "The wake is mentioned to define its spatial path relative to the vessel, consistent with the producer's 'anchor' classification.", "verdict": "anchor_reference" }, { "owned_object": "wheelhouse", "violating_phrase": "The wheelhouse, antenna, navigation light, and lifebuoy remain visible", "reason": "The wheelhouse is referenced as remaining visible on the hull, serving as an anchor in accordance with the producer's declaration.", "verdict": "anchor_reference" } ] } } ], "scene_type": "normal", "dependent_scene_index": -1, "dependency_reason": "", "_shot_index": 2, "visible_entities": [ "L18" ], "render_prompt_card": { "schema_version": 1, "shot_key": { "scene_index": 30, "shot_index": 2 }, "render_strategy": { "mode": "direct", "primary_subject": "", "framing_scale": "wide", "moment_lock": "single still moment from the selected shot only", "camera_direction": "From near water level, about 20 meters off the vessel's side and slightly aft of its bow, the camera tracks parallel as the bow crosses the sharply legible boundary of the dense fog band. The wide side view places open sea across the lower frame, the hull in the middle plane, and the fog bank ahead, preserving the physical moment when the bow disappears into it while the remaining hull is still visible outside.", "lighting_mood": "The setting sun casts a restrained golden reflection across the cool, desaturated sea while the dense fog remains tonally muted.", "perception_mode": "direct", "spatial_consistency": { "camera_frame_rule": { "applies_when": "camera_direction expresses a vertical camera height or viewpoint position 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, wide, insert} from `shot_staging.framing_scale` enum (render_strategy.framing_scale, production SOT). LLM consumes the enum only; keyword classification is forbidden.", "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" ], "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "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", "subject_reference_policy_cross_ref": "see id_policy.subject_reference_policy (subject_reference_policy SOT — policy enum 에 따라 ID/outlook 사용)" }, "self_check_steps": [ "read framing_scale enum from render_strategy.framing_scale (do NOT infer 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." }, "frame_spatial_contract": null, "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": [ "L18" ], "allowed_outlook_pairs": [], "must_use_composite_character_ids": true, "common_noun_required_when": [ "a reproduced face inside any declared reproduction surface", "unregistered extra" ], "id_use_required_when": [ "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon" ], "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" }, "reproduction_surface_rule": { "applies": true, "id_use": "forbidden — generic descriptor with demographic only", "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)." }, "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" }, "subject_reference_policy": [], "constraints": [ "ID format for visible subjects (C## bare, C##O## composite IDs, or common-noun descriptor) is determined per-subject by the id_policy per-subject policy array — see constraints[2] below", "do not use C##O## for a reproduced face inside any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a generic descriptor with demographic instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the generic-descriptor rule applies only to a face reproduction, not to the prop entity.", "per-subject identity reference policy is provided in id_policy.subject_reference_policy[] — apply each subject's policy when emitting references: (a) id_and_outlook_required ⇒ subject must use C##O## composite ID; (b) base_id_required ⇒ subject must use bare C## only (no outlook); (c) generic_descriptor_allowed ⇒ subject must NOT use C## or C##O##, render with a common/demographic descriptor only. omitted subjects default to id_and_outlook_required", "when an ID is required for a subject (policy = id_and_outlook_required or base_id_required) and a C##O## (or C##) is introduced 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. does not apply to generic_descriptor_allowed subjects (no ID emitted)", "when an ID is required (policy = id_and_outlook_required or base_id_required), 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). does not apply to generic_descriptor_allowed subjects — those use a common/demographic descriptor with no C##/C##O##" ] }, "background_binding": { "mode": "background_ref_attached", "bg_id": "L18B01", "reference_usage": "exact_background", "owned_objects": [ "antenna", "fishing boat", "fog bank", "lifebuoy", "navigation light", "sea", "wake", "wheelhouse" ], "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", "if focusing on an owned object, declare its reference-kind via the reference_phrase_kinds sidecar (producer contract)" ] }, "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'" ], "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy applies per-subject; policy enum (id_and_outlook_required / base_id_required / generic_descriptor_allowed) determines C##/C##O## use", "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. third-person full body and close-up of the same character must not be combined in a single prompt.", "third_person_handling": "describe persons with full or partial body in frame; single-camera consistency forbids combining a partial close-up and the same character's full body in one prompt", "close_up_handling": "an isolated framed area of a single subject must not coexist with that subject's full-body description in the same prompt", "focus_on_is_focus_area_not_view_switch": "any framing-intent wording designates the area shown by the single camera position; it does not switch the view", "subject_reference_policy_cross_ref": "id_policy.subject_reference_policy — subjects with policy = 'base_id_required' or 'generic_descriptor_allowed' must not use outlook reference (C##O## form); additionally, subjects with policy = 'generic_descriptor_allowed' must not use bare C## either — render with a common/demographic descriptor only", "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 — third-person full body and close-up of the same character must not be combined in a single shot. any framing-intent wording designates the area shown by the single camera position, not a view switch (id_policy.subject_reference_policy applies per-subject)", "props held by characters face one direction only — toward camera OR toward character, never both" ] }, "asset_requirements": { "required_refs": [ { "kind": "background", "id": "L18B01", "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" ] }, "render_contracts": [], "_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": "8a04963cc2ebe186" } ], "consumed_bg_catalog_hash": "8116c6d190dba2d1", "consumed_shot_binding_hash": "6b05a0058065215d" }, "step_id": "scene_detail", "run_id": "a800b014-2be2-4663-8572-559be6300f1e", "resolved_model": "gemini-pro", "updated_at": "2026-07-23T02:31:24.813608+00:00", "project_config_snapshot": {} }