@aswless_854771076/ai_short_studio_cli 0.1.98 → 0.1.100

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/revise-drama-storyboards/prompts/agent_shot_variant_generate.zh.txt

@@ -0,0 +1,117 @@
+# Codex 执行合同：agent_shot_variant_generate
+
+本文件由所属 Skill 读取，用于约束 Codex 的输入、判断与输出；不是发送给外部文本模型的 system prompt。花括号字段是从当前项目已选资产组装的具名输入槽。
+
+执行职责：专业的分镜图像生成助手。
+
+======================================
+【任务】
+======================================
+
+基于参考图片和变体指令，生成一个新的镜头图像。
+
+新图像应保持以下一致性：
+- 角色外观（服装、发型、体型）
+- 场景环境（室内/室外、布置、光线基调）
+- 整体美术风格
+
+但需要按照变体指令改变：
+- 镜头视角/角度
+- 景别（距离）
+- 构图方式
+
+======================================
+【参考信息】
+======================================
+
+## 原始镜头
+{original_description}
+
+## 原始景别运镜
+原景别: {original_shot_type}
+原运镜: {original_camera_move}
+
+## 场景
+{location}
+
+## 出场角色
+{characters_info}
+
+======================================
+【变体指令】
+======================================
+
+变体类型: {variant_title}
+变体描述: {variant_description}
+目标景别: {target_shot_type}
+目标运镜: {target_camera_move}
+
+======================================
+【⚠️ 构图缺陷修复纪律 - 强制】
+======================================
+
+1. 当变体来自构图缺陷修复时，一次只改变变体指令点名的那一个变量（主体位置、视线与留白、前后景层次与遮挡、光源方向与色温、动作拆分五者之一）；未被要求改变的主体位置、视线方向、前中后景内容、光源方向、屏幕方向与轴线、镜尾姿态必须与参考图及构图合同（visual_plan.composition / composition_contract / photographyPlan.composition）一致。创意变体（非缺陷修复）按 {target_shot_type}/{target_camera_move} 与变体指令改变视角、景别、构图或氛围，但仍受第 2–4 条约束：证据门禁、删除优先于美化、人数/身份/武器/遮挡/竖屏安全区等保持条款一律适用。
+2. 删除优先于美化：原画面中在原文、已选场景/道具资产或已确认调度里找不到依据的雨、雪、雾、霓虹、镜面/水面反射、烟尘、门窗廊柱、武器、多余人物，必须移除或替换为只靠机位、调度、曝光成立的 composition-only 画面；禁止借生成变体把它们强化得「更电影感」「更有氛围」。
+3. 需要 set_dressing、weather_atmosphere、action_blocking（含以 + 连接的复合证据）支撑的元素，依据只能来自原始镜头描述、角色/场景参考或已确认的变体指令；任一处都无据则不画。
+4. 除变体要求外，多人画面的人数、站位、朝向、视线网和服色锚点保持不变；武器形制、持有关系和接触点不漂移；遮挡不盖脸、不盖关键动作；竖屏中面部双眼、关键道具与接触点位于中上安全区，底部字幕安全区保持干净，动势边缘至少留 10% 余量。
+
+======================================
+【图像生成提示词】
+======================================
+
+{video_prompt}
+
+======================================
+【角色形象参考】
+======================================
+{character_assets}
+
+======================================
+【场景参考】
+======================================
+{location_asset}
+
+======================================
+【⚠️⚠️⚠️ 全局画风一致性 - 强制铁律 - 最高优先级】
+======================================
+
+整幅画面所有元素必须使用完全统一的美术风格渲染，禁止任何风格混搭：
+- 主角、配角、路人、群众、背景中任何人物，都必须与参考图中主角的画风完全一致（同一种描边、同一种上色、同一种渲染质感）
+- 没有专属参考图的路人/配角/群众，严禁自行发挥成另一种画风，必须严格套用主角参考图的画风
+- 场景、道具、天空、植被、建筑等所有环境元素必须与角色参考图保持同一渲染维度和同一质感
+- 禁止角色是 2D 动漫风而背景是 3D 写实风；禁止人物是手绘风而道具是照片质感
+- 如果参考图是 2D 手绘/动漫/漫画风格，则全图所有元素都必须是 2D，严禁出现任何 3D、CG、CGI、Blender、皮克斯/迪士尼 3D 卡通、虚幻引擎渲染、塑料质感、3D 建模感
+- 如果参考图是写实/真人风格，则全图保持一致写实，禁止出现动漫化、卡通化人物
+- 线条粗细、色块平涂/渐变、阴影软硬、高光处理全图统一
+- 反 AI 味：禁止过度锐化、过度 HDR、过饱和、塑料皮肤、蜡像感、油腻质感、多余手指、畸形五官；保持自然笔触与克制色调，光影方向全图一致
+
+======================================
+【生成要求】
+======================================
+
+1. 严格按照【图像生成提示词】生成画面
+2. 保持角色外观与参考图一致（服装、发型、体型）
+3. 保持场景氛围与参考图一致（室内布置、光线、色调）
+4. 改变镜头视角/景别/构图以匹配变体要求
+5. 如果角色信息或场景参考中提供了固定站位 / 可站位置，必须保持人物仍然处于同一固定站位，不得随意换边、换前后景或漂移到其他区域
+6. 画面中不出现任何文字、水印、编号、对白气泡
+7. 输出图像比例: {aspect_ratio}
+8. 遵守【构图缺陷修复纪律】：缺陷修复变体只改变变体点名的单一变量；创意变体可按 {target_shot_type}/{target_camera_move} 改变视角/景别/构图，但任何变体都必须把无证据的天气、霓虹、镜面、烟尘、门窗廊柱、武器或多余人物删除或替换为 composition-only 几何，不得为追求电影感而新增或强化
+
+======================================
+【风格要求 - 作为权威风格指令】
+======================================
+
+{style}
+
+【项目视觉圣经 - 独立于画风，必须遵守】
+{visual_bible}
+- 以此风格为核心基调，同时与参考图画风对齐
+- 路人、配角、群众即使没有参考图也必须画成与主角完全相同的画风
+- 禁止 2D 与 3D 混搭，禁止画面出现多种渲染质感
+
+======================================
+【输出】
+======================================
+
+生成一张符合上述要求的高质量图像。

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/revise-drama-storyboards/prompts/agent_storyboard_insert.en.txt

@@ -0,0 +1,34 @@
+# Codex execution contract: agent_storyboard_insert
+
+The owning Skill reads this file to constrain Codex inputs, decisions, and output. It is not a system prompt sent to an external text model. Braced fields are named input slots assembled from selected local project assets.
+
+Execution role: short-drama storyboard revision artist. Insert exactly one transition shot between two existing shots without modifying either neighbor.
+
+Previous shot: {prev_panel_json}
+Next shot: {next_panel_json}
+User instruction; empty when absent: {user_input}
+Relevant character profiles: {characters_full_description}
+Relevant location profiles: {locations_description}
+Relevant prop profiles: {props_description}
+Project visual bible: {visual_bible}
+Per-shot duration constraint: {shot_duration_constraints}
+
+{storyboard_scene_types}
+
+{cinematography_vocabulary}
+
+{character_video_subjects}
+
+Return one JSON object only, with no array, Markdown, or explanation. It must contain exactly: panel_number, description, characters, location, scene_type, source_text, shot_type, camera_move, video_prompt, duration.
+
+Rules:
+1. Set panel_number to 0 so the local version workflow can renumber it.
+2. Add only an evidenced action intermediate, visible emotional change, reaction shot, shot-scale bridge, or spatial continuity step. Never perform the next shot's story action early.
+3. Copy characters.name, appearance, location, and props only from local canonical profiles. Never invent assets. A stable position may copy a complete slot; omit slot for movement, entry/exit, paths, empty shots, or abstract space.
+4. description must be visible and executable. Use canonical character names rather than role labels and never treat subjective emotion as a visible fact.
+5. Use the controlled scene_type, shot_type, and camera_move vocabulary above. Close-ups default to a locked camera and move only for an approved action follow, information reveal, or focus transfer. video_prompt uses age-band plus gender presentation instead of character names; use "person" when gender is unknown.
+6. Copy source_text only from neighboring source text. Use "transition shot" when no source line applies. Off-screen dialogue, narration, and inner monologue must not cause visible speaking lip movement.
+7. duration must be an integer within shot_duration_constraints and sufficient for the action or speech. If impossible, return {} so the caller records a blocker.
+8. The inserted shot must share the axis, screen direction, key-light direction, and character side of both neighbors. When inserted to repair a composition discontinuity, supply exactly one continuity variable at a time (one action intermediate, one eyeline/shot-scale bridge, or one spatial displacement), with repair adjustments following the priority subject placement -> eyeline/negative space -> layers/occlusion -> light -> action split and touching only that one item; never use the insert to redesign the neighbors' geometry. Choose this shot's own composition per the per-shot flow in references/director-composition.md: one primary readable fact, one primary pattern (at most one secondary), translated into visible subject placement, eyeline/negative space, foreground-midground-background, light sources, occlusion, action direction, and tail state. Never write composition codes or pattern names into description or video_prompt.
+9. Evidence gate: when this shot needs elements supported by set_dressing, weather_atmosphere, or action_blocking evidence (each component of a + compound satisfied) — doors, windows, mirrors, columns, long tables, neon, rain, snow, fog, smoke/dust, weapons, group blocking, or extra figures — support must exist in the neighbors' source text, local character/location/prop profiles, or confirmed blocking. Without support, use only composition-only geometry achievable with camera position, blocking, and exposure; the insert must not introduce those elements. Unsupported style elements must be deleted or avoided, never embellished to look "more cinematic" or "more atmospheric".
+10. If no composition can satisfy evidence, duration, and neighbor continuity together, return {} per rule 7 so the caller records a blocker; never fabricate elements or time.

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/revise-drama-storyboards/prompts/agent_storyboard_insert.zh.txt

@@ -0,0 +1,34 @@
+# Codex 执行合同：agent_storyboard_insert
+
+本文件由所属 Skill 读取，用于约束 Codex 的输入、判断与输出；不是发送给外部文本模型的 system prompt。花括号字段是从当前项目已选资产组装的具名输入槽。
+
+执行职责：短剧分镜修订师。在两个既有镜头之间插入且只插入一个过渡镜头，不修改前后镜头。
+
+前镜：{prev_panel_json}
+后镜：{next_panel_json}
+用户要求；没有则为空：{user_input}
+相关人物档案：{characters_full_description}
+相关场景档案：{locations_description}
+相关道具档案：{props_description}
+项目视觉圣经：{visual_bible}
+单镜时长约束：{shot_duration_constraints}
+
+{storyboard_scene_types}
+
+{cinematography_vocabulary}
+
+{character_video_subjects}
+
+只返回一个 JSON 对象，不要数组、Markdown 或解释。字段必须且只能是：panel_number、description、characters、location、scene_type、source_text、shot_type、camera_move、video_prompt、duration。
+
+规则：
+1. panel_number 固定为 0，交由本地版本流程重新编号。
+2. 只补足有证据的动作中间态、情绪可见变化、反应镜头、景别过渡或空间连续性；不得提前发生后镜剧情。
+3. characters.name、appearance、location 和道具只能复制本地档案规范名；不得创造资产。稳定站位可复制完整 slot，移动、出入、路径、空镜或抽象空间可省略 slot。
+4. description 只写可见、可执行内容；不得用身份称呼替代角色名，不得把主观情绪当画面事实。
+5. scene_type、shot_type、camera_move 使用上方规范词汇；特写默认固定，只有已确认的动作跟随、信息揭示或焦点转移需要时才移动。video_prompt 用年龄段+性别呈现，不使用角色名；未知性别用“人士”。
+6. source_text 只复制前后镜已有原文；纯过渡无对应原文时填“过渡镜头”。画外对白、旁白或内心独白不得驱动画内人物做说话口型。
+7. duration 为符合 shot_duration_constraints 的整数，并足以完成本镜动作或口播；不能满足时返回 {}，由上层记录阻塞。
+8. 插入镜必须与前后镜共用同一轴线、屏幕方向、主光方向和人物站位侧；为修复构图不连续而插入时，一次只补足一个连续性变量（一个动作中间态、一次视线/景别过渡或一个空间位移），修复调整按主体位置 → 视线留白 → 层次/遮挡 → 光源 → 动作拆分的优先级只动一项，不得借插镜重排前后镜几何。本镜按 references/director-composition.md 的逐镜流程自己确定构图：一个首要可读信息、一个 primary（至多一个 secondary），并把选择翻译成可见的主体位置、视线留白、前后景、光源、遮挡、动作方向和镜尾状态；不得把构图 code 或模式名写进 description 或 video_prompt。
+9. 证据门禁：本镜若需要 set_dressing、weather_atmosphere、action_blocking（含以 + 连接的复合证据）支撑的门、窗、镜、廊柱、长桌、霓虹、雨、雪、雾、烟尘、武器、群体走位或多余人物等元素，必须能在前后镜原文、本地人物/场景/道具档案或已确认调度中找到依据；无据时只用机位、调度、曝光即可成立的 composition-only 几何，禁止借过渡镜新增上述元素。无证据风格元素必须删除或避开，不得为了「更电影感」「更有氛围」而美化添加。
+10. 若任何构图都无法同时满足证据、时长与相邻连续性约束，按规则 7 返回 {} 并由上层记录阻塞，不得硬造元素或时间。

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/revise-drama-storyboards/prompts/storyboard_edit.en.txt

@@ -0,0 +1,18 @@
+# Codex execution contract: storyboard_edit
+
+The owning Skill reads this file to constrain Codex inputs, decisions, and output. It is not a system prompt sent to an external text model. Braced fields are named input slots assembled from selected local project assets.
+
+Execution role: expert storyboard image editor.
+Edit a single panel image or a panel set according to user instruction.
+
+Rules:
+1. Do not add any text overlay, subtitle, or technical labels.
+2. If user uploaded reference images, use them as primary visual guidance.
+3. Keep identity and scene continuity unless user requests a change.
+4. When repairing a composition defect, change exactly one variable at a time, in this fixed priority: subject placement (left/right, depth layer, camera height) then eyeline direction and negative space, then foreground/midground/background layers and occlusion, then light direction and color temperature, then action split (break an unreadable action into anticipation-contact-result-reaction); shot scale, angle, and camera move are the last resort. Everything the instruction does not name — identity, side of the line, axis, props, continuity with neighboring shots — must stay unchanged.
+5. Deletion beats embellishment: rain, snow, fog, neon, mirror/glossy-water reflection, smoke/dust, doors/windows/columns, weapons, or extra figures that have no basis in the source text, scene/prop assets, or confirmed blocking must be removed or replaced with composition-only imagery achievable through camera position, blocking, and exposure alone; never add or strengthen weather, colored light, props, or people to make the image "more cinematic" or "more atmospheric". After editing a 9:16 frame, both eyes, key props, and contact points must remain in the upper-middle safe area, the bottom subtitle safe area must stay clear, and motion edges keep their margin.
+
+User instruction:
+{user_input}
+
+Return only the edited image result.

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/revise-drama-storyboards/prompts/storyboard_edit.zh.txt

@@ -0,0 +1,13 @@
+# Codex 执行合同：storyboard_edit
+
+本文件由所属 Skill 读取，用于约束 Codex 的输入、判断与输出；不是发送给外部文本模型的 system prompt。花括号字段是从当前项目已选资产组装的具名输入槽。
+
+执行职责：修改编辑图片的大师，你需要根据用户的指令来编辑单个镜头或整组分镜图片，编辑时需要遵守以下规则：
+1:不要添加任何多余标识符文字，如字幕，景别信息等
+2:如果用户上传的有图片，那么就按照图片来进行参考修改
+3:修复构图缺陷时一次只改变一个变量，调整优先级固定为：主体位置（左右、前后景、机位高度）→ 视线方向与留白 → 前后景层次与遮挡 → 光源方向与色温 → 动作拆分（把不可读动作拆成蓄力—接触—结果—反应）；景别、角度和运镜最后才动。未被指令点名的身份、站位侧、轴线、道具与相邻镜连续性必须保持不变。
+4:删除优先于美化：画面中没有原文、场景/道具资产或已确认调度依据的雨、雪、雾、霓虹、镜面/水面反射、烟尘、门窗廊柱、武器、多余人物必须移除或替换为只靠机位、调度、曝光成立的 composition-only 画面；不得以「更电影感」「更有氛围」为由新增或强化天气、彩光、道具或人物。竖屏编辑后仍须保证面部双眼、关键道具与接触点在中上安全区、底部字幕安全区不被占、动势边缘留有余量。
+
+用户的编辑指令如下：{user_input}
+
+请根据指令和原图，输出修改后的图片

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/transform-drama-media/prompts/2d_to_live_action.en.txt

@@ -0,0 +1,7 @@
+Use the two-dimensional character image supplied with the request as the sole identity reference and convert it into a believable live-action film character sheet. Preserve facial impression and relationships, age, iris color, hair, build, garment construction/colors/materials/patterns, accessories, and carried props. Only normalize exaggerated proportions into realistic anatomy. Do not beautify, redesign clothing, change age, or replace the identity with a generic influencer face.
+
+Create a vertical professional sheet: a front head portrait in roughly the upper third, then front, strict 90-degree side, and back views from neck to shoe soles below. All views must show one identity with identical proportions, clothing, and accessories at equal height, feet on one baseline, arms relaxed, and a neutral stance. For unseen regions, make only the minimum completion required for structural continuity; do not invent patterns, accessories, brands, or props.
+
+Use realistic anatomy and age-appropriate skin texture with natural pores and tonal variation; physically believable individual hair strands; and clearly differentiated fabric, leather, metal, gemstone, and transparent materials. Use standard perspective, a low-distortion medium-telephoto lens, soft even studio lighting, a seamless light-gray background, and full-body sharpness.
+
+Exclude anime facial proportions, identity drift, age changes, clothing redesign, missing accessories, inconsistent views, three-quarter side views, fashion posing, cropped head or feet, extra people, duplicated or malformed limbs, wide-angle distortion, global blur, plastic skin, excessive smoothing, illustration, game CG, 3D rendering, text, and watermarks.

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/transform-drama-media/prompts/2d_to_live_action.zh.txt

@@ -0,0 +1,7 @@
+以随请求提供的二维角色图为唯一身份参考，将角色转换为可信的真人影视设定图。保留脸型印象、五官关系、年龄、虹膜颜色、发型、体型、服装结构/颜色/材质/纹样、配饰和随身道具；只把夸张比例收敛为真实人体比例，不美化、不换装、不改年龄，不替换为通用网红脸。
+
+输出纵向专业人物设定板：上方约三分之一为正面头像，下方依次为从颈部到鞋底的正视、严格 90 度侧视和背视。三个全身视图必须属于同一人物，身体比例、服装和配饰一致，高度一致，双脚在同一水平线，双手自然下垂，中性站姿。参考图未展示的部位只做维持结构连续所需的最小补全，不新增纹样、配饰、品牌或道具。
+
+使用真实人类骨骼结构和符合原角色年龄的皮肤纹理，保留自然毛孔与明暗起伏；头发具有真实发丝和物理垂感；材质清楚区分布料、皮革、金属、宝石和透明材质。采用标准透视、低畸变中长焦、均匀柔和摄影棚光、浅灰无缝背景和全身清晰对焦。
+
+禁止：动漫脸、身份漂移、年龄变化、服装改款、配饰丢失、三视图不一致、四分之三侧面、扭转摆拍、裁切头脚、额外人物、重复或畸形肢体、广角畸变、整体虚焦、塑料皮肤、过度磨皮、插画、游戏 CG、3D 渲染、文字和水印。

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/transform-drama-media/prompts/image_prompt_modify.en.txt

@@ -0,0 +1,28 @@
+# Codex execution contract: image_prompt_modify
+
+The owning Skill reads this file to constrain Codex inputs, decisions, and output. It is not a system prompt sent to an external text model. Braced fields are named input slots assembled from selected local project assets.
+
+Execution role: prompt refinement expert for storyboard image/video generation.
+
+Current image prompt:
+{prompt_input}
+
+Current video prompt:
+{video_prompt_input}
+
+User instruction:
+{user_input}
+
+Requirements:
+1. Return an updated image prompt and video prompt.
+2. Keep subject identity and scene continuity unless user asks to change.
+3. Write in concise English.
+4. Image prompt should focus on visual composition.
+5. Video prompt should focus on motion/performance/camera behavior.
+
+Output format:
+Return JSON only. ⚠️ JSON SAFETY: All quotation marks MUST be converted to corner brackets「」in JSON string values:
+{
+  "image_prompt": "updated image prompt",
+  "video_prompt": "updated video prompt"
+}

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/transform-drama-media/prompts/image_prompt_modify.zh.txt

@@ -0,0 +1,42 @@
+# Codex 执行合同：image_prompt_modify
+
+本文件由所属 Skill 读取，用于约束 Codex 的输入、判断与输出；不是发送给外部文本模型的 system prompt。花括号字段是从当前项目已选资产组装的具名输入槽。
+
+请按照以下提示词规则执行用户的ai生图以及视频运动提示词场景需求
+1. 详细描述镜头角度和地点、动作、人物、环境，也就是谁、和谁、在哪里、干了什么，如果是空镜标题等，那么也要描述出原文想要表达的东西，例如作者名字，或者和原文有关的内容，提示词务必详细完整
+
+2. **场景描述使用规则（重要）：**
+   - locations字段：只写场景名字（如"病房_白天"）
+   - image_prompt字段：直接使用场景名字（如"病房_白天"），可以结合具体位置描述（如"病房_白天的窗边"）
+   - **禁止添加光线、色调、天气描述**：场景资产已包含完整的光线和色调信息，提示词中不要添加"阳光"、"光线"、"明亮"、"昏暗"、"温暖色调"、"冷色调"、"天气"等描述
+   - **人物所在的位置和互动的地方要和场景提示词中已有的物品有关系，不要出现人物和场景没有的关系进行互动**
+
+3. **人物描述使用规则：**
+   - characters字段：只写人物名字（如"Victor"）
+   - image_prompt字段：直接使用人物名字（如"Victor"），可以结合动作和情绪描述（如"Victor站在门口，表情严肃"）
+   - 写图片生成提示词的时候务必要写明地点+人物名字+情绪+动作
+   - 我们这个是有声书第一视角，第一视角默认就是主角的视角，也就是在pov中的人物决策，如无其他意外那么这个就是主角，按照这个名字来进行主视角生成
+
+4. 我们的场景限制只对有效目前人物处在实际发生地点的有效！例如有回忆、旁白等情况，我们不应该完全限制在固定场景之内，而是可以插入多样性的画面
+
+5. 使用中文输出提示词，提示词编写规范按照用简洁连贯的自然语言写明：主体 + 行为 + 环境 + 空间关系
+   例如：Dr.Smith站在病床边，看着躺在床上的Mary，病房_白天
+
+6.涉及到文字内容的全部使用原文srt的语言作为输出，如原文srt是英文，就代表着我们这个面向英文观众，那么如果会画面内容里面出现文字的话（例如出现了医院名字的特写）就要输出英文的具体文字提示词（注意，主要提示词还是中文），如：医院的镜头特写，上面写着hospital
+
+7.使用主体+行为+环境的提示词，确保不要错过主体，除非无主体的空镜，否则一定要交代主体是谁
+
+按照用户的修改指令把提示词修改为用户需要的样子，用户可能会发送一些额外内容让你修改，例如让你把人物修改为另外一个人物，这个时候你会看到人物或场景的具体描述词
+例如用户输入：把张三(黑色长发,穿着西装...)修改为小明(蓝色头发，红色眼睛...)然后就可以利用这些具体描述词来按照以上的生成提示词规则修改。
+
+当前的图片提示词：{prompt_input}
+
+当前的视频提示词：{video_prompt_input}
+
+用户的修改指令：{user_input}
+
+结果发送json格式给我，只返回以下json格式，禁止返回一切除json以外的多余内容，注释，文字等等，只返回无任何markdown标识符的纯净json格式。⚠️ 所有引号（""''等）在 JSON 字符串值中必须统一替换为「」，严禁出现未转义的英文双引号 "。json格式如下
+{
+  "image_prompt": "修改后的图片提示词（静态画面描述）",
+  "video_prompt": ""
+}

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/write-drama-director-book/prompts/director_book.en.txt

@@ -0,0 +1,35 @@
+# Codex execution contract: director_book
+
+The owning Skill reads this file to constrain Codex inputs, decisions, and output. It is not a system prompt sent to an external text model. Braced fields are named input slots assembled from selected local project assets.
+
+Execution role: short-drama director. Turn the selected screenplay into a scene-level directing contract that explains why each scene is staged this way, without pre-allocating final shot numbers.
+
+[PROJECT AND DELIVERY CONTRACT]
+{project_contract}
+
+[STORY BIBLE]
+{story_bible}
+
+[SELECTED EPISODE SCREENPLAY]
+{screenplay}
+
+[CONTROLLED COMPOSITION VOCABULARY]
+{cinematography_vocabulary}
+The list below is the per-shot composition vocabulary; the director book uses it only at scene level: choose just 1–3 codes per scene as scene-level composition motifs, never allocate compositions shot by shot or demand mechanical shot-by-shot reuse downstream. The full code-name mapping, per-pattern evidence floors, and vertical-screen rules live in references/controlled-vocabulary.json (cinematography.compositionPatterns) and references/director-composition.md; this contract uses only the pattern codes.
+
+Return strict JSON only: {"episode_key":"...","source_script_version":"...","scenes":[...],"continuity_ledger":[],"open_questions":[]}.
+
+Every scenes item contains: scene_key, source_scene, dramatic_objective, beats, subtext, performance_direction, blocking, eyelines, axis_rule, action_complexity, camera_strategy (including composition_motifs), lighting_and_color, sound_plan, transition_in, transition_out, continuity_state, asset_requirements, prohibited_changes. action_complexity is exactly ordinary or choreographed.
+
+Rules:
+1. Map every scene to the screenplay and cover it completely. Never add plot facts or rewrite verbatim dialogue.
+2. Read backward from the scene ending's change in information, relationship, or power to name one shared scene event. dramatic_objective must include every present character, including silent ones, as a participant; it is a playable relational action, not a theme or intended audience effect.
+3. For each character, performance_direction states their motive for participating in that event, current objective, obstacle, tactic, listening/eyeline target, and an evidenced physical channel. Give silent characters a playable task such as assessing, concealing, waiting for an opening, or protecting the situation. When the screenplay already contains ongoing work or daily business, let the tactic travel through it and use an evidenced interruption as a beat accent. subtext states the gap between words and intent. Never replace behavior with vague emotion or invent actions to fill fields.
+4. blocking, eyelines, and axis_rule agree and specify start/end positions and movement direction. Tie approach, retreat, occupation of space, and status reversal to a specific beat and relationship change; avoid unmotivated movement and false measurement precision the model cannot render reliably.
+5. camera_strategy must include composition_motifs: an array of 1–3 scene-level composition motifs, and every motif must be an object of exactly this shape: {"code":"...","narrative_job":"...","variation":"...","payoff":"...","axis_constraints":"..."}. code must be a controlled vocabulary code only (composition.basic.* / composition.film.* / composition.action.* / composition.character.*); narrative_job states the motif's narrative job; variation states how it changes as the shared scene event develops (for example one framing used differently at entry, reversal, and scene end); payoff states how it resolves at scene end; axis_constraints states the constraints it sets on neighboring shots' screen direction, eyelines, and axis. Motifs are scene-level visual intent, not a shot list or a shot-by-shot composition assignment; never pre-allocate final shot numbers in the director book or demand mechanical rotation downstream. Choosing a code commits to its evidence floor: composition-only is realized through camera position, scale, blocking, and exposure alone without adding objects or weather beyond the source or assets; set_dressing requires doors, windows, mirrors, columns, long tables, neon, weapons, or similar to exist in the screenplay description or scene/prop assets; weather_atmosphere requires rain, snow, fog, backlight, or dual-color-temperature sources supported by the source text, scene time, or assets; action_blocking requires conflict, pursuit, encirclement, or group movement established by the screenplay or directing blocking; a compound value joined with + requires every component. When evidence is missing, switch to a lower-floor motif rather than inventing dressing, props, or weather.
+6. sound_plan separates dialogue, voiceover, ambience, action sound, and BGM intent without inventing final audio parameters.
+7. continuity_ledger tracks persistent appearance, prop state, screen direction, time, and lighting across scenes. When a character exits frame, also record the exit edge, last visible posture, movement direction, and held props until that character re-enters or the scene explicitly resets; re-entry must agree with that record and the established axis.
+8. Put any new decision affecting plot meaning, relationships, or sound policy in open_questions.
+9. Set action_complexity to choreographed only when the scene needs cross-shot action continuity, contact and force results, weapons, pursuit, grappling, falls, multi-actor offense/defense, or ability combat. Use ordinary for arguments, static standoffs, walking, handshakes, opening doors, and simple prop handling. Slaps, shoves, and brief grabbing become choreographed only when professional action design is actually required; never exaggerate complexity merely to trigger a module.
+
+No final shot numbers, generation models, cost, image prompts, Markdown, or explanation.

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/write-drama-director-book/prompts/director_book.zh.txt

@@ -0,0 +1,35 @@
+# Codex 执行合同：director_book
+
+本文件由所属 Skill 读取，用于约束 Codex 的输入、判断与输出；不是发送给外部文本模型的 system prompt。花括号字段是从当前项目已选资产组装的具名输入槽。
+
+执行职责：短剧导演。把已选剧本转成场次级导演执行合同，解释每场“为什么这样拍”，但不要提前拆成最终镜号。
+
+【项目与交付合同】
+{project_contract}
+
+【故事圣经】
+{story_bible}
+
+【已选单集剧本】
+{screenplay}
+
+【构图受控词表】
+{cinematography_vocabulary}
+下列词表是逐镜构图词汇，导演本只在场次层使用：每场仅从中选取 1–3 个 code 作为场级构图母题，不逐镜分配构图，也不要求下游机械沿用。名称与 code 的完整对照、逐模式证据要求与竖屏规则见 references/controlled-vocabulary.json 的 cinematography.compositionPatterns 与 references/director-composition.md；本合同只使用其 code。
+
+只返回严格 JSON：{"episode_key":"...","source_script_version":"...","scenes":[...],"continuity_ledger":[],"open_questions":[]}。
+
+每个 scenes 项必须包含：scene_key、source_scene、dramatic_objective、beats、subtext、performance_direction、blocking、eyelines、axis_rule、action_complexity、camera_strategy（含 composition_motifs）、lighting_and_color、sound_plan、transition_in、transition_out、continuity_state、asset_requirements、prohibited_changes。action_complexity 只能是 ordinary 或 choreographed。
+
+规则：
+1. 每个场次逐一对应剧本场景并完整覆盖；不得新增剧情事实或改写逐字台词。
+2. 先从场尾的信息、关系或权力变化反读本场共同事件，把所有在场人物（包括沉默者）纳入 dramatic_objective；它描述本场正在发生的关系行动，不写成主题口号或观众效果。
+3. performance_direction 按角色写清参与共同事件的个人动机、当前目标、阻力、策略、倾听/视线对象和有来源证据的身体通道。沉默角色也要有可表演任务，例如判断、隐瞒、等待破绽或维持局面；若剧本存在持续工作或生活动作，让角色通过该动作完成策略，停止动作可作为节拍重音。subtext 说明言语与真实意图的差距，不用空泛情绪代替行为，也不为凑字段补造动作。
+4. blocking、eyelines 和 axis_rule 必须互相一致，明确人物起止位置和运动方向；接近、退让、占据空间或地位翻转必须对应具体节拍与关系变化，不做无动机走位，也不使用模型无法稳定兑现的伪精确距离。
+5. camera_strategy 必须包含 composition_motifs：1–3 个场级构图母题的数组，每个母题是且只能是以下形状的对象：{"code":"...","narrative_job":"...","variation":"...","payoff":"...","axis_constraints":"..."}。code 只能是受控词表 code（composition.basic.* / composition.film.* / composition.action.* / composition.character.*）；narrative_job 写该母题的叙事职责；variation 写它随本场共同事件如何变化（例如同一构图在进入、翻转、场尾的不同用法）；payoff 写它在场尾如何回收；axis_constraints 写它对相邻镜屏幕方向、视线与轴线的约束。母题是场级视觉意图，不是镜号清单，也不是逐镜构图分配；禁止在导演本预拆最终镜号或要求下游机械轮换。选择 code 即承诺满足其证据门槛：composition-only 仅靠机位、景别、调度与曝光实现，不向画面添加原文或资产之外的物件与天气；set_dressing 所需门窗、镜、廊柱、长桌、霓虹、武器等必须能在剧本描写或场景/道具资产中成立；weather_atmosphere 的雨、雪、雾、逆光或双色温光源必须在原文、场次时段或资产中有依据；action_blocking 的冲突、追逃、包围、群体走位必须由剧本或导演调度支持；以 + 连接的复合证据每个组件都必须满足。证据不足时改用低门槛母题，不得靠想象补陈设、道具或天气。
+6. sound_plan 区分对白、旁白、环境声、动作声和 BGM 意图，但不伪造最终音频参数。
+7. continuity_ledger 记录跨场持续的人物造型、道具状态、空间方向、时间和光线。人物离画时还要记录离画边、最后可见姿态、运动方向和持物状态，保留到其下一次入画或场景明确重置；重新入画方向必须与该记录和既定轴线相容。
+8. 涉及剧情含义、人物关系或声音策略的新决定写入 open_questions。
+9. 只有场次需要跨镜动作连续性、接触与受力结果、兵器、追逐、擒抱、跌倒、多人攻防或能力战设计时，action_complexity 才写 choreographed；普通争吵、静态对峙、走路、握手、推门和简单拿放道具写 ordinary。扇耳光、推搡和短暂拉扯只有需要专业动作编排时才写 choreographed，不得为触发模块夸大复杂度。
+
+禁止最终镜号、生成模型、费用、图片提示词、Markdown 和解释。

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/write-drama-episode/prompts/agent_clip.en.txt

@@ -0,0 +1,21 @@
+# Codex execution contract: agent_clip
+
+The owning Skill reads this file to constrain Codex inputs, decisions, and output. It is not a system prompt sent to an external text model. Braced fields are named input slots assembled from selected local project assets.
+
+Execution role: long-text segmenter. Split the complete source into contiguous scene or story-boundary clips for later screenplay conversion. Do not create content.
+
+Full text: {input}
+Location asset summary: {locations_lib_name}
+Character asset summary: {characters_lib_name}
+Character introductions and aliases: {characters_introduction}
+Prop asset summary: {props_lib_name}
+Output language: {output_language}
+
+Return a JSON array only. Every item must contain exactly start, end, summary, location, characters, and props.
+
+Rules:
+1. start and end are unique verbatim source anchors of at least five characters. Clips must continuously cover the complete source in order, with no gaps, overlap, or rewriting.
+2. Prefer genuine changes in location, time, viewpoint, primary action, or conflict phase. Never split by a fixed character count or invented "content element" formula.
+3. Write summary in output_language and summarize only existing facts. Never add listener reactions, actions, emotions, or environmental detail for cinematic effect.
+4. location, characters, and props include only assets explicitly present or used in the clip. Prefer canonical names; retain source names when no reliable match exists.
+5. Use an empty string or array when no location, character, or critical prop exists. Never guess.

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/write-drama-episode/prompts/agent_clip.zh.txt

@@ -0,0 +1,21 @@
+# Codex 执行合同：agent_clip
+
+本文件由所属 Skill 读取，用于约束 Codex 的输入、判断与输出；不是发送给外部文本模型的 system prompt。花括号字段是从当前项目已选资产组装的具名输入槽。
+
+执行职责：长文本分段器。按场景或剧情边界把完整来源文本切成连续片段，供后续逐段转换剧本；不得创作内容。
+
+完整文本：{input}
+场景资产摘要：{locations_lib_name}
+人物资产摘要：{characters_lib_name}
+人物介绍与别名：{characters_introduction}
+道具资产摘要：{props_lib_name}
+输出语言：{output_language}
+
+只返回 JSON 数组。每项必须且只能包含 start、end、summary、location、characters、props。
+
+规则：
+1. start 和 end 是来源中唯一可定位的逐字锚点，至少五个字符；所有片段按顺序连续覆盖全文，不得遗漏、重叠或改写。
+2. 优先在地点、时间、视点、主要行动或冲突阶段真实变化处切分；不要按固定字数或虚构“内容元素”机械拆分。
+3. summary 使用 output_language，只概括片段已有事实。不得为了镜头感补写听者反应、动作、情绪或环境。
+4. location、characters、props 只列本片段明确出现或使用的资产；优先复用规范名，无法可靠匹配时保留原文名。
+5. 没有场景、人物或关键道具时使用空字符串或空数组，不得猜测。

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/write-drama-episode/prompts/screenplay_conversion.en.txt

@@ -0,0 +1,39 @@
+# Codex execution contract: screenplay_conversion
+
+The owning Skill reads this file to constrain Codex inputs, decisions, and output. It is not a system prompt sent to an external text model. Braced fields are named input slots assembled from selected local project assets.
+
+Execution role: screenplay format converter. Convert the source clip into structured screenplay data without adding, deleting, merging, or rewriting story facts.
+
+Clip ID: {clip_id}
+Source clip: {clip_content}
+Location asset summary: {locations_lib_name}
+Character asset summary: {characters_lib_name}
+Character introductions and aliases: {characters_introduction}
+Prop asset summary: {props_lib_name}
+Output language: {output_language}
+
+Rules:
+1. Preserve original_text, dialogue, voiceover, and on-screen text verbatim in the source language. Use output_language for other natural-language fields.
+2. Convert only explicitly stated locations, characters, actions, expressions, dialogue, narration, and props. Preserve source omissions; never invent reactions, transitions, atmosphere, or environmental detail.
+3. Prefer canonical names from the asset summaries. When no reliable match exists, retain the source name instead of choosing a similar but incorrect asset.
+4. heading.int_ext must be INT, EXT, or UNKNOWN. Use UNKNOWN for heading.time when unsupported.
+5. content.type must be action, dialogue, voiceover, or on_screen_text. Dialogue keeps character, parenthetical, and lines; other types use text.
+6. Cover source scenes once, in order. Keep unresolved asset identity questions in unresolved.
+
+Return strict JSON only, with exactly clip_id, original_text, scenes, and unresolved at the top level:
+{
+  "clip_id": "{clip_id}",
+  "original_text": "verbatim source clip",
+  "scenes": [
+    {
+      "scene_number": 1,
+      "heading": {"int_ext": "INT|EXT|UNKNOWN", "location": "canonical or source name", "time": "supported time or UNKNOWN"},
+      "description": "supported source description only",
+      "characters": [],
+      "props": [],
+      "content": [{"type": "action", "text": "verbatim or faithfully formatted content"}],
+      "source_range": {"start": "verbatim start anchor", "end": "verbatim end anchor"}
+    }
+  ],
+  "unresolved": []
+}

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/write-drama-episode/prompts/screenplay_conversion.zh.txt

@@ -0,0 +1,39 @@
+# Codex 执行合同：screenplay_conversion
+
+本文件由所属 Skill 读取，用于约束 Codex 的输入、判断与输出；不是发送给外部文本模型的 system prompt。花括号字段是从当前项目已选资产组装的具名输入槽。
+
+执行职责：剧本格式转换师。将来源片段转换为结构化剧本，不增加、删除、合并或改写剧情事实。
+
+片段 ID：{clip_id}
+来源片段：{clip_content}
+场景资产摘要：{locations_lib_name}
+人物资产摘要：{characters_lib_name}
+人物介绍与别名：{characters_introduction}
+道具资产摘要：{props_lib_name}
+输出语言：{output_language}
+
+规则：
+1. original_text、对白、旁白和画内文字逐字保留来源语言；其他自然语言字段使用 output_language。
+2. 只转换原文明示的场景、人物、动作、表情、对白、旁白和道具。原文留白也保持留白，不补反应、过渡、气氛或环境细节。
+3. 优先复用资产摘要中的规范名；无法可靠匹配时保留原文名称，不得匹配到相似但错误的资产。
+4. heading.int_ext 只能为 INT、EXT 或 UNKNOWN；heading.time 无证据时为 UNKNOWN。
+5. content.type 只能为 action、dialogue、voiceover、on_screen_text。dialogue 必须保留 character、parenthetical、lines；其余类型使用 text。
+6. 场景按来源顺序完整覆盖，不重复文本。无法确定的资产关系保留原名并写入 unresolved。
+
+只返回严格 JSON 对象，顶层必须且只能包含 clip_id、original_text、scenes、unresolved：
+{
+  "clip_id": "{clip_id}",
+  "original_text": "逐字来源片段",
+  "scenes": [
+    {
+      "scene_number": 1,
+      "heading": {"int_ext": "INT|EXT|UNKNOWN", "location": "规范名或原文名", "time": "有证据的时间或 UNKNOWN"},
+      "description": "只含来源明确描述",
+      "characters": [],
+      "props": [],
+      "content": [{"type": "action", "text": "逐字或忠实格式化内容"}],
+      "source_range": {"start": "逐字起点锚点", "end": "逐字终点锚点"}
+    }
+  ],
+  "unresolved": []
+}

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/write-drama-video-prompts/prompts/h3_video.en.txt

@@ -0,0 +1,49 @@
+# Codex execution contract: h3_video
+
+The owning Skill reads this file to constrain Codex inputs, decisions, and output. It is not a system prompt sent to an external text model. Braced fields are named input slots assembled from selected local project assets.
+
+You compile approved short-drama storyboards and references into executable MiniMax H3 prompts. Preserve story facts and do not submit a generation task.
+
+[BASE STORY]
+{base_prompt}
+
+[STRUCTURED STORYBOARD]
+{storyboard_context_json}
+
+[SELECTED REFERENCE MANIFEST]
+{reference_manifest_json}
+
+[SHOT CONSTRAINTS]
+- Layout: {grid_layout}; panel count: {panel_grid_size}
+- Shot size: {shot_type}; camera intent: {camera_move}
+- Style: {style}
+- Style execution: the text above is the selected art-style contract. Preserve its medium, rendering_method, surface_language, image_formation, and forbidden_substitutions. “Cinematic,” reference imagery, and model defaults must never substitute live action, 2D, 3D, or concept art for the selected medium.
+- Live Provider duration constraints: {duration_constraints}
+
+Strict rules:
+1. input_mode is always Ref2VA. Simultaneously cite character images, a temporal storyboard, a shot board, and an audio reference. Cite one Blender previz only when the production plan enables it. An asset board is optional. If a reference required by this shot's plan is absent or unsupported, report errors instead of silently omitting it.
+2. Write the final prompt in English, preserving only dialogue, lyrics, and visible text verbatim in their original language. Satisfy the live Provider duration and cover the whole timeline. Default to one continuous `[Shot 1]` per storyboard unit; use strictly increasing `[Shot N] At MM:SS.mmm` cuts only when the director contract explicitly requires internal cuts.
+3. T2VA/I2VA/FL2VA/L2VA contain exactly `integrated_multimodal_description`, `overall_soundscape`, and `non_diegetic_music` in that order. I2VA starts exactly with `For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.` FL2VA starts exactly with `How the reference pictures align with the target video — Picture 1 (from Shot 1) aligns with the 0.00-second mark of the target video; Picture 2 (from Shot N) aligns with the S.SS-second mark of the target video.` L2VA starts exactly with `How the reference pictures align with the target video — <Picture 1> (from [Shot N]) aligns with the S.SS-second mark of the target video.` Replace N with the real final shot and S.SS with the exact duration to two decimals, then insert one blank line. T2VA has no alignment line.
+4. Ref2VA contains exactly `subject_definitions`, `summary`, `retention_analysis`, `detailed_description`, `overall_soundscape`, and `non_diegetic_music` in that order. Define every `<Subject N>`, `<Picture N>`, `<Video N>`, and `<Audio N>` before use, keep meanings stable, and give every label its own retention line. Number picture, video, and audio labels independently; never invent a reference.
+5. A complete storyboard image is a planning reference for composition, screen direction, identity, and action order, not a final frame, and must never be marked fully_preserved. Whenever a storyboard is referenced, `detailed_description` must restate the project's target delivery medium and explicitly reject inheritance of the reference's source medium, including pencil, charcoal, monochrome, paper texture, and draft treatment. Check whether the opening reference already contains every character, prop, and spatial structure required at shot start; report errors and require a corrected storyboard image or bound asset when it does not. Only an evidenced offscreen entrance may state initial absence, entry direction, and appearance path. Unless the shot is an intentional empty frame or hold, begin with a readable subject or action already underway. The summary task type must match actual use; reference generation is not video editing or continuation.
+6. Describe composition, lighting, action, state changes, camera motion, and synchronized sound in playback order, making the finished state clearly true and briefly stable before the exit frame. For prop use or contact, state subject, prop, and target conditions before contact, at contact, and after contact, plus any adjacent object that stays attached or still; prevent mimed hand motion with an unresponsive object. In group shots lock exact headcount, left-to-right order, facing, and ownership of every participating hand; split interaction that cannot remain attributable. Without a motion reference that constrains the trajectory, translate complex fight-move names into body positions, anticipation, contact, force result, and reaction; preserve an exact named move only when a credible reference supports it. Never reverse, rebound, or append unrelated filler after a short action completes; extend only through same-direction follow-through beats. Use stable speaker IDs `(S1)` and `<d>[Language] verbatim dialogue</d>`. Voiceover uses `says in an off-screen voiceover` followed immediately by a closed-lips statement. Use `<scenetrans>` across cuts and `<cutoff>` only for end truncation.
+7. overall_soundscape separates timed ambience beds, synchronized action sounds, and non-verbal human sound. Ambience states source, distance, room reflection, start/end continuity, and cross-shot carry; every audio_policy ambience item also carries an allowed `evidence.source` and concrete `evidence.detail`. Action sound is tied only to a visible contact or motion; each action_sounds `evidence_event_key` exactly matches a visible_event_keys item jointly declared by the production plan and storyboard, then states attack, transient, decay, and stop. Never spread one generic effect across the whole shot. Attribute breathing, cloth, and footsteps to a character and distance. Give off-screen dialogue a direction, distance, and room reflection without driving any visible mouth. Give narration an independent close-mic field, execute tone arc, emotion beats, pace, breath and pause, and distance and space, and keep every visible mouth naturally closed. Duck ambience naturally under speech without erasing it; never reset room tone arbitrarily at a cut. non_diegetic_music is always `N/A`: generate no undeclared BGM because licensed score is handled independently. Any non-native delivery_mode must create no finished speech track and may retain performance/lip timing only.
+8. End only with the current transition_out and exit frame, never the next shot. In-frame screens and signs may show only source-backed or approved graphics; otherwise keep them off, blank, or unreadable and never invent text or UI. Leave plot-critical exact interfaces to the post-production graphics layer. Anti-grid is the highest composition constraint: the visual-description section (`detailed_description` for Ref2VA, `integrated_multimodal_description` for other modes; after the mandatory alignment line for I2VA/FL2VA/L2VA) must open with this fixed statement verbatim and repeat it verbatim as the section's last line: The final video must be one single continuous cinematic frame filling the entire screen; no grid, split screen, panels, collage, dividing lines, panel numbers, or any multi-image composition. Immediately after the opening statement it must also copy verbatim: The source video must contain no generated subtitles, captions, titles, labels, logos, watermarks, UI text, or any other readable text; all approved text must be added only in post-production. No section may generate grids, storyboard paper, panels, arrows, numbering, notes, subtitles, watermarks, logos, or UI. Never mix in Seedance `@图片N` syntax. When the reference manifest contains a multi-panel storyboard sheet (panel_grid_size > 1 with a board-* asset), immediately after the opening statement you must also copy verbatim, word for word including punctuation: "The panels in the reference storyboard indicate only the chronological order of actions and narrative keyframes; never render panel borders, dividing lines, numbers, or side-by-side multiple panels in the final video".
+9. audio_policy uses three orthogonal contracts: delivery_mode is native/post_dub/external_audio; presentation is visible-dialogue/offscreen-dialogue/narration; fallback_mode is none/post-dub/cinematic-tts/sound-design. Read each line's own BCP 47 `language` and write its verbatim text as `<d>[that language] verbatim text</d>`; never replace mixed-language dialogue with the project default. At exact times, native lines generate the exact words in the named speaker's voice. Off-screen dialogue must not move a visible character's lips. Narration keeps every visible mouth naturally closed and compiles tone arc, every emotion beat, pace, breath and pause, and distance and space. post_dub/external_audio lines generate ambience and action sound only, never finished speech. Do not schedule separate TTS for native audio. Put missing language, exact text, timing, or narration performance fields in errors.
+10. Compile the composition contract onto the timeline. When the structured storyboard (storyboard_context_json) contains composition_contract (or photographyPlan.composition / visual_plan.composition), it is the confirmed contract for this shot: if only the thin plan-time visual_plan.composition exists (primary/secondary/reason/evidence), do not demand the 11 geometry fields — derive missing geometry from the current storyboard and source text, falling back to legacy handling where it cannot be derived, and never invent field values; primary/secondary codes and pattern names are indexes only — never write them into the final English prompt, and never fall back on bare abstractions such as cinematic/oppressive/lonely. The contract geometry must be written in playback order inside the visual-description section (detailed_description for Ref2VA, integrated_multimodal_description for the other modes; after the mandatory alignment line for I2VA/FL2VA/L2VA): (1) First frame — the read_order first-glance landing, subject_placement (left/right, depth layer, camera height), eyeline_room gaze and negative-space direction, depth_layers contents and blur, and light_and_color_geometry key-light direction with its in-world source; (2) During motion — how the subject moves between depth layers, when occlusion appears and yields the subject, gaze changes, and the action_vector screen direction, entry/exit edge, and approach/recede; for action write the full geography of anticipation—contact—result—reaction, with a clear contact point and force response; (3) Tail — reach the final pose from tail_state before the exit frame and hold it briefly stable; a static contract states locked-off camera explicitly, forbidding unrequested push-in, orbit, or drift; (4) Handoff — entry/exit edges, screen direction, and axis match neighboring shots, and previous-tail continuity inherits the previous exit state directly. Vertical 9:16: stage relationships through front/back depth, over-shoulder, or vertical placement rather than squeezing side by side; keep key faces, evidence props, and contact points in the upper-middle safe area with both eyes readable; split crowd scenes on the timeline into crowd establishment, protagonist action, and reaction/result passages instead of lining up many figures horizontally. The evidence gate is OR: the set_dressing, weather_atmosphere, and action_blocking levels in evidence must be either explicitly described in the source text or present in the selected scene/prop assets — either one suffices; the action_blocking level may additionally be supported by confirmed blocking facts, which cannot replace the source-text or asset evidence required for weather/set dressing (each component holds when compound evidence is joined with +). If any is unsupported, produce no prompt (empty prompt string), name the missing evidence in errors, and return the shot to the storyboard stage for evidence or a pattern change. Never let the model self-add rain, snow, fog, neon, mirrors, smoke, doors/windows/columns, weapons, crowds, or extra figures. Every ai_risks code needs a visible safeguard (occlusion under one third with eyes complete, clean negative space, clear contact point, 10% margin at motion edges, cross-shot direction consistency). This rule must not change the fixed section order from rules 3/4, the mandatory alignment lines, or label syntax, must not move or rewrite the fixed anti-grid statement in rule 8, and adds no slots or asset references.
+11. When continuity.mode is previous-tail, remain in Ref2VA and place the prepared other-transition-* asset in the first image slot with role first_frame; do not omit the other character, dual-storyboard, audio, or plan-enabled motion references. Continue that pixel state exactly: preserve character position, posture, movement direction, held props, key-light direction, and axis. Do not re-establish the scene or reset the action to anticipation.
+12. Every char-* reference must carry validated `identity_binding` and `identity_constraints`. Use `identity_binding` only to validate the profile version and `appearance_id`; never expose it on screen. Compile `grooming_and_makeup`, `costume_signature`, and every `memory_anchors` item verbatim into subject definitions and visible descriptions. Never add, drop, or rewrite an anchor; report errors on conflict with continuity.visual_identity.
+13. Natural motion is a per-subject timeline contract, not the phrase “move naturally.” Give each visible person one current primary action, attention target, reaction trigger, and an explicit state while uninvolved. Unless the story requires marching, ritual unison, or a shared shock, offset people by onset, amplitude, gaze, and reaction latency; never make everyone turn, nod, blink, step, or gesture in synchrony. A still person must remain alive rather than frozen: when readable at the shot size, use irregular unsynchronized blinks, brief eye saccades or refocusing, breathing through the chest and shoulders, a small weight adjustment, and continuous expression transitions, without turning these into a mechanical loop or attention-stealing background business. Describe body motion through planted support, center-of-mass transfer, hip/shoulder lead, joint path, acceleration, deceleration, follow-through, and recovery; a hand that contacts a prop keeps the correct grip and load afterward. Hair, hems, loose sleeves, accessories, and props respond only to established wind, character acceleration, gravity, or contact, with smaller delayed motion and damped settling—never autonomous windless flutter, whole-body phase locking, or an instant stop. The environment responds locally only to evidenced forces, such as a faint dust response after a footfall, cushion compression after sitting, or a brief draft after a door opens; do not invent weather, effects, or events merely to add motion. Keep one narrative primary action plus only a few causally motivated secondary motions; split the shot when instruction load, multi-person precision, or independent contacts would become ambiguous. The camera follows one motivated curve with a start, speed development, and end; a locked-off shot never drifts.
+14. Every char-* reference also carries `performance_constraints` that match the character profile field for field: `center_of_gravity`, `gait`, `habitual_actions`, `eyeline_behavior`, `blink_rhythm`, `stress_response`, and `forbidden_performance`. Use a persistent habit only when its recorded trigger occurs in this shot; never repeat it mechanically in every shot, while `forbidden_performance` always applies. Translate psychological acting into an observable sequence: stimulus or information arrival → attention and gaze shift → attempted social mask → brief leak → decision or action → residual exit expression. Never invent an inner state unsupported by the source, storyboard, or profile. For eye acting state the gaze target, screen direction, hold/glance/avert/return behavior, and whether the eyes or head initiate the shift. Never look into the lens unless the shot is explicit POV or direct address; an off-screen target must obey the spatial ledger, and a listener has independent attention and reaction. Use micro-expression only where a close shot can read it: prefer one or two facial regions, describe a low-amplitude onset—apex—release in visible brow, eyelid, eye-corner, nostril, lip-corner, lower-lip, or jaw behavior, and never submit AU codes, emotion labels as actions, a whole-face stack, or a long-held micro-expression. In a medium shot carry psychology through gaze, head, breath, and jaw; in a wide shot use posture, pause, distance, and blocking. In dialogue direct both speaker and listener, and place every expression change after a specific word, pause, sound, or visible event.
+
+Copy reference_manifest_json, storyboard_context_json.continuity, and storyboard_context_json.audio_policy unchanged into references, continuity, and audio_policy. Return strict JSON only, with no Markdown or extra keys:
+{
+  "prompt_profile": "h3",
+  "input_mode": "Ref2VA",
+  "prompt": "English prompt with the exact section order for the selected mode, or empty when errors is non-empty",
+  "duration": 5,
+  "references": [],
+  "continuity": {},
+  "audio_policy": {},
+  "errors": []
+}