create-storyboard

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Create Storyboard

创建分镜

Mission

任务

Turn a script into a SceneDance/Seedance-ready production package that behaves like it was prepared by a director, storyboard artist, editor, and AI video production coordinator. The output is not a list of visual descriptions; it is a continuity-controlled plan for generating many

0-15s

video clips that can be cut together smoothly.

Always optimize for:

film continuity between independently generated clips
one clear action chain and one main camera movement per SceneDance generation
explicit action start/end and emotion start/end for every shot
deliberate shot handoffs: each clip ending plants a visual, spatial, motion, or sound clue that the next clip receives
reference-image discipline: character, scene, prop, keyframe, and storyboard inputs
handoff and edit-point design before prompt writing
post-production usability in Jianying/CapCut

将脚本转化为可直接用于 SceneDance/Seedance 的制作包，其完成度应如同由导演、分镜师、剪辑师和 AI 视频制作协调员共同准备。输出并非一组视觉描述，而是一份受连续性控制的规划，用于生成许多可以流畅剪辑组合的

0-15s

视频片段。

始终优先优化：

独立生成片段之间的电影连续性
每次 SceneDance 生成只包含一条清晰动作链和一种主要镜头运动
每个镜头明确动作开始/结束和情绪开始/结束
明确定义镜头交接：每个片段结尾植入视觉、空间、运动或声音线索，供下一片段承接
参考图规范：角色、场景、道具、关键帧和分镜输入
在编写提示词前完成交接和剪辑点设计
后期制作对剪映/CapCut 的易用性

Runtime Assumptions

运行时假设

The bundled scaffold script is deterministic and can run without an AI model once Python is available.
Production storyboard planning needs a Codex-style agent that can read/write files, inspect scripts and generated artifacts, run the scaffold script, and revise production documents.
Use a multimodal model when the task includes reference images, character sheets, keyframes, storyboard boards, or generated video/image review.
Use a strong reasoning model for real production packages. The author-known-good profile is macOS, Codex-style local agent, GPT-5.5-class reasoning, and
```
xhigh
```
reasoning for complex packages.
Use
```
high
```
reasoning for normal production work; use
```
xhigh
```
when available for long-form stories, multi-scene continuity, or dense handoff/edit matrices.
Smaller models may run the scaffold but are more likely to weaken continuity, handoff logic, bilingual prompt consistency, deterministic board assembly, or QA decisions.
For macOS/Windows/Linux setup details, read
```
references/runtime-and-platform.md
```
before promising reproducibility to a user.

随附的脚手架脚本是确定性的；安装 Python 后即可在没有 AI 模型的情况下运行。
制作分镜规划需要 Codex-style agent，能够读写文件、检查脚本和生成产物、运行脚手架脚本并修订制作文档。
当任务包含参考图、角色表、关键帧、分镜板或生成的视频/图像审查时，使用多模态模型。
实际制作包应使用强大的推理模型。作者验证有效的配置是 macOS、Codex-style 本地 agent、GPT-5.5 级推理能力，并在复杂制作包中使用
```
xhigh
```
推理。
常规制作工作使用
```
high
```
推理；对于长篇故事、多场景连续性或密集的交接/剪辑矩阵，在可用时使用
```
xhigh
```
。
较小的模型可以运行脚手架，但更可能削弱连续性、交接逻辑、中英双语提示词一致性、确定性分镜板拼装或 QA 决策。
在向用户承诺可复现性之前，请先阅读
```
references/runtime-and-platform.md
```
，了解 macOS/Windows/Linux 的设置详情。

Hard Rules

硬性规则

SceneDance/Seedance clips must be
```
<= 15s
```
.
New projects default to
```
SH### = CLIP###
```
: one film shot equals one SceneDance video generation. A
```
CLIP###
```
may cover multiple shots only for low-risk inserts or when the user explicitly asks.
A user request for "分镜图", "重新生成分镜图", "storyboard images", or "SceneDance inputs" means per-clip deliverables by default. Do not satisfy it with one full-film overview board unless the user explicitly asks for an overview/contact sheet.
A full-film overview board, master storyboard sheet, or contact sheet is only for review. It is not a final SceneDance storyboard image and must not be counted as delivered
```
CLIP###
```
storyboard output.
Every final storyboard image must cover exactly one SceneDance generation unit:
```
final_image_package/clip_storyboards/<CLIP###>_storyboard_<time-range>.png
```
.
Final storyboard boards must be production boards, not AI-generated mood boards. Do not ask an image model to generate the final board layout, labels, captions, tables, or readable production text.
Image models may generate only clean visual sources for storyboard boards: start panel, key-action panel, edit-out panel, handoff panel, or clean keyframes. Assemble the final storyboard board with deterministic local layout code, HTML/CSS screenshot, a design tool, or another controllable renderer.
A final storyboard board must contain readable, deterministic production metadata:
```
CLIP ID
```
, time range, duration, scene/location, aspect ratio, tone/style,
```
START
```
,
```
KEY ACTION
```
,
```
EDIT OUT
```
, camera method, action start/end, emotion start/end, handoff-out, edit boundary, audio bridge, risk/fallback, and reference-image combination.
Reject and regenerate/rebuild any storyboard board that has empty caption areas, missing
```
CLIP ID
```
, missing time range, missing start/key/edit-out structure, unreadable/garbled metadata, mixed multiple clips, poster-like composition, or lacks edit/handoff information.
Durations are story-driven. Do not split by fixed totals such as 4 x 15s for a minute. Use
```
2-4s
```
for inserts/reactions,
```
4-7s
```
for clear physical actions,
```
8-12s
```
for sustained performance or atmosphere, and
```
12-15s
```
only for stable long takes.
Every shot must define: purpose, duration, shot size, camera movement, composition, character state, action start/end, emotion start/end, receiver-in state, handoff-out state, motion vector, spatial bridge, occlusion carrier, visual bridge, reference images, SceneDance prompt, previous transition, next transition, edit note, risk, and fallback.
Build the continuity bible before writing final prompts. Lock identity, wardrobe, hair, props, scene geography, 180-degree axis, eyelines, screen direction, light, weather, time state, color, aspect ratio, lens language, and key object positions.
Design the handoff matrix and edit boundary matrix before image generation. Every neighboring pair needs what the prior clip hands off, what the next clip receives, spatial entrance/exit, motion direction, occlusion carrier, visual bridge, audio bridge, edit type, frame-match requirement, CapCut handling, and fallback cut.
Default boundary strategy is editable continuity, not strict frame continuity. Prefer action match, eyeline match, screen-direction match, composition match, cutaway, insert, reaction, empty-room shot, occlusion cut, hard cut, J-cut, or L-cut. Use strict end-frame/start-frame matching only when the action truly must remain continuous.
A boundary cannot be labeled only as "natural cut", "smooth transition", or "hard cut" unless the handoff design states the baton being passed, or the shot explicitly chooses a deliberate jump cut/emotional rupture.
Do not ask SceneDance to solve complex multi-person blocking, multiple camera cuts, or too many action beats inside one generation. Split with close-ups, inserts, reactions, props, or atmosphere shots.
Camera movement must be chosen for story and SceneDance stability, not repeated by habit. Use static shots only when motivated; consider tracking, lateral move, foreground occlusion push, pull-back reveal, handheld micro-move, POV, over-shoulder, low/high angle, door-frame peek, prop-led move, light-led move, or UI foreground occlusion when they serve the handoff.
A storyboard board is not the main SceneDance input if it contains text/grid/multiple cells. The primary video input should be a clean keyframe from
```
final_image_package/clip_keyframes/
```
or
```
05_images/selected/
```
.
If two neighboring shots are merged because their total duration is
```
<=15s
```
, they become one explicit
```
CLIP###
```
in
```
clip_plan.md
```
. Generate one clean start keyframe and one single-clip production storyboard board for that merged clip; do not feed SceneDance a two-panel or multi-shot board as the primary image.
If the user has not specified aspect ratio, ask one concise question before final prompts or image generation. If target duration is not inferable, ask before final shot planning.

SceneDance/Seedance 片段必须
```
<= 15s
```
。
新项目默认采用
```
SH### = CLIP###
```
：一个电影镜头对应一次 SceneDance 视频生成。只有在低风险插入镜头或用户明确要求时，一个
```
CLIP###
```
才能涵盖多个镜头。
用户要求“分镜图”“重新生成分镜图”“storyboard images”或“SceneDance inputs”时，默认需要按片段分别交付。除非用户明确要求总览图/联系表，否则不能仅用一张全片总览板来满足要求。
全片总览板、主分镜表或联系表仅用于审阅。它们不是最终的 SceneDance 分镜图，也不得计为已交付的
```
CLIP###
```
分镜图输出。
每张最终分镜图必须且只能覆盖一个 SceneDance 生成单元：
```
final_image_package/clip_storyboards/<CLIP###>_storyboard_<time-range>.png
```
。
最终分镜板必须是制作板，而不是 AI 生成的情绪板。不得要求图像模型生成最终分镜板的布局、标签、字幕、表格或可读制作文字。
图像模型只能为分镜板生成干净的视觉素材源，包括起始画面板、关键动作画面板、剪出画面板、交接画面板或干净关键帧。最终分镜板必须使用确定性的本地布局代码、HTML/CSS 截图、设计工具或其他可控渲染器进行拼装。
最终分镜板必须包含可读且确定性的制作元数据：
```
CLIP ID
```
、时间范围、时长、场景/地点、宽高比、基调/风格、
```
START
```
、
```
KEY ACTION
```
、
```
EDIT OUT
```
、镜头方法、动作开始/结束、情绪开始/结束、交接输出、剪辑边界、音频桥接、风险/回退方案和参考图组合。
如果分镜板存在空白字幕区域、缺少
```
CLIP ID
```
、缺少时间范围、缺少起始/关键动作/剪出结构、元数据无法阅读或乱码、混合多个片段、呈现海报式构图，或缺少剪辑/交接信息，则必须拒绝并重新生成或重建该分镜板。
时长应由故事驱动。不得按每分钟固定拆分为 4 x 15s。插入镜头/反应镜头使用
```
2-4s
```
，清晰的身体动作使用
```
4-7s
```
，持续表演或氛围使用
```
8-12s
```
，只有稳定的长镜头才使用
```
12-15s
```
。
每个镜头必须定义：目的、时长、景别、镜头运动、构图、角色状态、动作开始/结束、情绪开始/结束、接收状态、交接输出状态、运动向量、空间桥接、遮挡载体、视觉桥接、参考图、SceneDance 提示词、上一转场、下一转场、剪辑备注、风险和回退方案。
在编写最终提示词前，先建立连续性圣经。锁定身份、服装、发型、道具、场景地理关系、180 度轴线、视线、银幕运动方向、光线、天气、时间状态、色彩、宽高比、镜头语言和关键物体位置。
在图像生成前设计交接矩阵和剪辑边界矩阵。每一对相邻镜头都需要明确前一片段交接什么、下一片段接收什么、空间进入/退出方式、运动方向、遮挡载体、视觉桥接、音频桥接、剪辑类型、帧匹配要求、CapCut 处理方式和备用切点。
默认边界策略是“可编辑连续性”，而不是严格的逐帧连续性。优先使用 action match、eyeline match、screen-direction match、composition match、cutaway、insert、reaction、empty-room shot、occlusion cut、hard cut、J-cut 或 L-cut。只有当动作确实必须连续时，才使用严格的尾帧/首帧匹配。
除非交接设计明确说明正在传递的“交接棒”，或者镜头有意选择 jump cut/情绪断裂，否则边界不能仅标记为“natural cut”“smooth transition”或“hard cut”。
不得要求 SceneDance 在一次生成中处理复杂的多人走位、多次镜头切换或过多动作节拍。应使用特写、插入镜头、反应镜头、道具镜头或氛围镜头进行拆分。
镜头运动应根据故事和 SceneDance 稳定性来选择，不能形成习惯性重复。只有在存在合理动机时才使用固定镜头；还可考虑 tracking、lateral move、foreground occlusion push、pull-back reveal、handheld micro-move、POV、over-shoulder、low/high angle、door-frame peek、prop-led move、light-led move 或 UI foreground occlusion，以服务于交接。
如果分镜板包含文字、网格或多个画面格，则不能将其作为 SceneDance 的主要视频输入。主要视频输入应为
```
final_image_package/clip_keyframes/
```
或
```
05_images/selected/
```
中的干净关键帧。
如果两个相邻镜头因总时长
```
<=15s
```
而被合并，则它们会在
```
clip_plan.md
```
中成为一个明确的
```
CLIP###
```
。为该合并片段生成一张干净的起始关键帧和一张单片段制作分镜板；不得将双画面或多镜头分镜板作为 SceneDance 的主要输入图像。
如果用户尚未指定宽高比，应在编写最终提示词或生成图像前简洁地询问一次。如果无法推断目标时长，也应先询问。

Standard Workflow

标准工作流

Extract the brief: video type, audience/platform, target duration, aspect ratio, story intent, tone, characters/products, locations, props, dialogue/audio, and deliverable scope.
Analyze the script into scenes and dramatic beats. Identify emotional turns, physical actions, object interactions, reveals, and places where a shot can hand off the next space or hide generation discontinuity.

Create bibles:

```
character_bible.md
```
```
scene_bible.md
```
```
product_prop_bible.md
```
```
style_bible.md
```
```
continuity_bible.md
```

Build the asset plan: character turnarounds, expression sheets, pose sheets, scene establishing/reverse angles, prop/product sheets, clean start keyframes, key-action frames, edit-out frames, optional bridge frames, and final clip storyboard boards.
Create shot cards with one
```
SH###
```
per default
```
CLIP###
```
. For every shot, choose duration by action load, emotion, information density, cut rhythm, and handoff requirement.
Write the reference input matrix. Each SceneDance shot must list the primary clean input image and all auxiliary character/scene/prop/storyboard references.
Build
```
handoff_design_matrix.md
```
for every neighboring shot before final prompts. The previous clip must plant the next clip's space, motion, visual token, or sound cue; the next clip must begin by receiving it.
Build the edit boundary matrix from the handoff matrix. Use continuity editing: action match, eyeline match, screen direction, composition/rhythm match, shot-size progression, reaction, insert, cutaway, occlusion, J-cut, and L-cut.
Write Image 2 prompts in separate Chinese and English files. Do not mix languages in the same generation prompt.
Generate or prepare reference images first when image generation is requested. Generate clean visual panels/keyframes first, then assemble one deterministic production storyboard board per
```
CLIP###
```
. Do not use an AI-generated board layout as final output, and do not generate a full-film overview board unless explicitly requested as an extra review image.
Write SceneDance shot prompts: selected image, duration, receiver-in state, action start/end, emotion start/end, camera movement, continuity locks, handoff-out state, edit-out visual token, next-scene clue, edit handles, audio bridge, and avoid list.
Write post-edit materials: SceneDance usage list, edit continuity notes, Jianying/CapCut edit plan, risk/fallback plan, and image manifest.
Validate: every clip is
```
<=15s
```
, every shot has a handoff plan and boundary plan, every shot has a primary input image ID, every promised image path is tracked, and no recurring identity/space/style lock is missing.

提取简报：视频类型、受众/平台、目标时长、宽高比、故事意图、基调、角色/产品、场景、道具、对白/音频和交付范围。
将脚本分析为场景和戏剧节拍。识别情绪转折、身体动作、物体互动、揭示，以及可用于交接下一空间或隐藏生成不连续性的镜头位置。

创建各类圣经：

```
character_bible.md
```
```
scene_bible.md
```
```
product_prop_bible.md
```
```
style_bible.md
```
```
continuity_bible.md
```

构建资产计划：角色多角度图、表情表、姿势表、场景定机位/反向机位、道具/产品表、干净起始关键帧、关键动作帧、剪出帧、可选桥接帧和最终片段分镜板。
创建镜头卡，默认每个
```
CLIP###
```
对应一个
```
SH###
```
。根据动作负荷、情绪、信息密度、剪辑节奏和交接需求确定每个镜头的时长。
编写参考输入矩阵。每个 SceneDance 镜头都必须列出主要干净输入图像，以及所有辅助角色/场景/道具/分镜参考图。
在编写最终提示词前，为每个相邻镜头构建
```
handoff_design_matrix.md
```
。前一片段必须为下一片段植入空间、运动、视觉标记或声音线索；下一片段必须通过承接该线索开始。
根据交接矩阵构建剪辑边界矩阵。使用连续性剪辑：action match、eyeline match、screen direction、composition/rhythm match、shot-size progression、reaction、insert、cutaway、occlusion、J-cut 和 L-cut。
将 Image 2 提示词分别写入中文和英文文件。同一生成提示词中不得混用语言。
当用户要求图像生成时，先生成或准备参考图。先生成干净视觉画面板/关键帧，然后为每个
```
CLIP###
```
拼装一张确定性的制作分镜板。不得将 AI 生成的版面作为最终输出，除非用户明确要求额外提供全片总览审阅图。
编写 SceneDance 镜头提示词：选定图像、时长、接收状态、动作开始/结束、情绪开始/结束、镜头运动、连续性锁定项、交接输出状态、剪出视觉标记、下一场景线索、剪辑余量、音频桥接和避免项。
编写后期制作材料：SceneDance 使用清单、剪辑连续性备注、剪映/CapCut 剪辑计划、风险/回退计划和图像清单。
验证：每个片段均
```
<=15s
```
，每个镜头都有交接计划和边界计划，每个镜头都有主要输入图像 ID，每项承诺交付的图像路径均有记录，并且不存在遗漏的重复身份/空间/风格锁定项。

Production Package

制作包

Use the scaffold script for new packages:

bash

python3 ~/.codex/skills/create-storyboard/scripts/create_storyboard_package.py <project-slug> --root <workspace-root> --title "<title>" --duration "<target-duration>" --aspect "<aspect-ratio>"

Omit

--aspect

only during planning. Final prompts and image generation require a confirmed aspect ratio.

When this skill folder is installed by itself, run the bundled smoke demo from the skill root:

bash

python3 scripts/create_storyboard_package.py demo-storyboard \
  --root outputs \
  --title "Demo Storyboard" \
  --duration "30s" \
  --aspect "16:9"

The package contains:

text

storyboard_projects/<project-slug>/
├── 01_script_brief/
├── 02_bibles/
├── 03_storyboard/
├── 04_prompts/
├── 05_images/
├── 06_delivery/
└── final_image_package/

For exact files and fields, read

assets/production_package_spec.md

. For the fillable production template, read

assets/storyboard_template.md

. For prompt structures, read

assets/img2_seedance_prompt_template.md

. For the detailed workflow and continuity/editing rules, read

references/storyboard_workflow.md

使用脚手架脚本创建新制作包：

bash

python3 ~/.codex/skills/create-storyboard/scripts/create_storyboard_package.py <project-slug> --root <workspace-root> --title "<title>" --duration "<target-duration>" --aspect "<aspect-ratio>"

仅在规划阶段省略

--aspect

。最终提示词和图像生成需要确认的宽高比。

单独安装此 skill 文件夹后，从 skill 根目录运行随附的冒烟演示：

bash

python3 scripts/create_storyboard_package.py demo-storyboard \
  --root outputs \
  --title "Demo Storyboard" \
  --duration "30s" \
  --aspect "16:9"

制作包包含：

text

storyboard_projects/<project-slug>/
├── 01_script_brief/
├── 02_bibles/
├── 03_storyboard/
├── 04_prompts/
├── 05_images/
├── 06_delivery/
└── final_image_package/

如需了解准确的文件和字段，请阅读

assets/production_package_spec.md

。如需可填写的制作模板，请阅读

assets/storyboard_template.md

。如需提示词结构，请阅读

assets/img2_seedance_prompt_template.md

。如需详细工作流及连续性/剪辑规则，请阅读

references/storyboard_workflow.md

。

Required Outputs

必需输出

A complete production package must include:

script analysis and project brief
character, scene, prop/product, style, and continuity bibles
asset generation list
master storyboard and detailed shot cards
SceneDance reference input matrix
handoff design matrix
edit boundary matrix
Image 2 Chinese prompts and English prompts
SceneDance shot prompts
SceneDance usage list
post-edit/Jianying/CapCut plan
risk and fallback plan
final image manifest

完整的制作包必须包括：

脚本分析和项目简报
角色、场景、道具/产品、风格和连续性圣经
资产生成清单
主分镜和详细镜头卡
SceneDance 参考输入矩阵
交接设计矩阵
剪辑边界矩阵
Image 2 中文提示词和英文提示词
SceneDance 镜头提示词
SceneDance 使用清单
后期剪辑/剪映/CapCut 计划
风险和回退计划
最终图像清单

Shot Card Schema

镜头卡 Schema

Every shot card must be human-readable Markdown and include a YAML block with these required keys:

yaml

shot_id: SH001
clip_id: CLIP001
scene_id: S001
purpose: ""
duration: ""
shot_size: ""
camera_movement: ""
composition: ""
character_state: ""
action_start: ""
action_end: ""
emotion_start: ""
emotion_end: ""
receiver_in: ""
handoff_out: ""
motion_vector: ""
spatial_bridge: ""
occlusion_carrier: ""
visual_bridge: ""
handoff_risk_reduction: ""
reference_images: []
scenedance_prompt: ""
prev_transition: ""
next_transition: ""
edit_notes: ""
risks: []
fallback_plan: ""

每张镜头卡都必须采用便于人工阅读的 Markdown 格式，并包含一个具有下列必需键的 YAML 块：

yaml

shot_id: SH001
clip_id: CLIP001
scene_id: S001
purpose: ""
duration: ""
shot_size: ""
camera_movement: ""
composition: ""
character_state: ""
action_start: ""
action_end: ""
emotion_start: ""
emotion_end: ""
receiver_in: ""
handoff_out: ""
motion_vector: ""
spatial_bridge: ""
occlusion_carrier: ""
visual_bridge: ""
handoff_risk_reduction: ""
reference_images: []
scenedance_prompt: ""
prev_transition: ""
next_transition: ""
edit_notes: ""
risks: []
fallback_plan: ""

Editing Logic To Apply

需要应用的剪辑逻辑

Use film language deliberately:

Establish geography before relying on eyelines or movement direction.
Respect the 180-degree axis unless the shot card explicitly designs an axis reset.
Treat every neighboring pair as a baton pass: the prior clip's ending must offer a receiver object, motion, foreground, light, color, doorway, UI layer, sound, or spatial clue that the next clip can inherit.
Do not rely on AI interpolation to invent continuity between unrelated images. Design the video itself: camera movement, foreground occlusion, composition extension, spatial entrances/exits, and sound carry-over.
Use eyeline matches: a character looks off-screen, then cut to what they see.
Use action matches: a hand reaches, then cut to the prop close-up; a head turns, then cut to the reaction or POV.
Use screen-direction matches: entering/exiting left/right must stay meaningful across space.
Use shot-size rhythm: wide to medium to close-up for orientation, action, emotion; close-up to insert for detail; reaction shot to absorb discontinuity.
Use inserts, props, empty rooms, occlusion, foreground wipes, door frames, passing vehicles, darkness, flashes, or motion blur to hide AI discontinuity.
Use J-cuts and L-cuts: let dialogue, ambience, music, footsteps, object sounds, or impact sounds bridge across clips.
Leave
```
0.5-1s
```
edit handles when possible, so generated starts/ends can be trimmed.

有意识地运用电影语言：

在依赖视线或运动方向之前，先建立地理空间关系。
除非镜头卡明确设计轴线重置，否则应遵循 180 度轴线。
将每一对相邻镜头视为一次交接棒传递：前一片段的结尾必须提供下一片段可以继承的承接对象、运动、前景、光线、颜色、门洞、UI 层、声音或空间线索。
不要依赖 AI 插帧在无关图像之间虚构连续性。应直接设计视频：镜头运动、前景遮挡、构图延展、空间进出和声音延续。
使用 eyeline match：角色看向画外，随后切至其所看到的内容。
使用 action match：手伸出去，随后切至道具特写；人物转头，随后切至反应镜头或 POV。
使用 screen-direction match：从左/右进入或离开的动作必须在空间关系中保持一致且有意义。
使用 shot-size rhythm：全景到中景再到特写，用于交代环境、动作和情绪；特写到插入镜头用于展示细节；使用反应镜头吸收生成不连续性。
使用插入镜头、道具、空房间、遮挡、前景擦除、门框、驶过的车辆、黑暗、闪光或运动模糊来隐藏 AI 不连续性。
使用 J-cut 和 L-cut：让对白、环境声、音乐、脚步、物体声音或撞击声跨片段延续。
尽可能预留
```
0.5-1s
```
的剪辑余量，以便裁剪生成片段的开头和结尾。

Camera Language Library

镜头语言库

Choose one main camera method per clip and state why it serves the shot or handoff:

```
locked-off
```
: stable observation, product clarity, visual contrast, or precise insert.
```
slow push-in
```
: emotional pressure, reveal, or attention narrowing; avoid using as the default for every shot.
```
pull-back reveal
```
: reveal a new space, hidden object, crowd, UI state, or consequence.
```
lateral track
```
: follow movement direction, hand off screen-left/screen-right geography, or pass behind foreground.
```
following track
```
: walk-with-character, corridor/doorway movement, entering a new space.
```
foreground occlusion push
```
: let a door frame, body, shelf, sign, smoke, rain, vehicle, or UI layer wipe the frame into the next clip.
```
POV / subjective
```
: receive an eyeline and show what the character sees.
```
over-shoulder
```
: preserve dialogue axis and spatial relation.
```
low/high angle
```
: emphasize power, vulnerability, scale, or product hero status.
```
handheld micro-move
```
: tension and human presence; keep motion small for SceneDance stability.
```
prop-led / light-led move
```
: let a held object, screen glow, flashlight, product reflection, or color field pull the viewer into the next shot.

每个片段选择一种主要镜头方法，并说明它为何服务于该镜头或交接：

```
locked-off
```
：稳定观察、产品清晰呈现、视觉对比或精确插入镜头。
```
slow push-in
```
：情绪压力、揭示或收窄注意力；不要将其作为每个镜头的默认选择。
```
pull-back reveal
```
：揭示新空间、隐藏物体、人群、UI 状态或后果。
```
lateral track
```
：跟随运动方向、交接画面左/右侧空间关系，或从前景后方穿过。
```
following track
```
：与角色同行、穿过走廊/门洞或进入新空间。
```
foreground occlusion push
```
：让门框、身体、货架、标牌、烟雾、雨、车辆或 UI 层擦过画面并连接到下一片段。
```
POV / subjective
```
：承接角色视线并展示角色看到的内容。
```
over-shoulder
```
：保持对白轴线和空间关系。
```
low/high angle
```
：强调权力、脆弱感、尺度或产品英雄地位。
```
handheld micro-move
```
：营造紧张感和人 Presence；保持动作轻微，以确保 SceneDance 稳定性。
```
prop-led / light-led move
```
：让手持物体、屏幕光、 flashlight、产品反光或色域将观众视线引向下一镜头。

Image Generation Handling

图像生成处理

When generating images:

Generate Chinese-prompt images into
```
05_images/zh/
```
.
Generate English-prompt images into
```
05_images/en/
```
.
Put selected clean SceneDance inputs into
```
05_images/selected/
```
.
Put final clean clip keyframes into
```
final_image_package/clip_keyframes/
```
.
Put generated clean storyboard panel sources into
```
final_image_package/clip_storyboards/panels/
```
or another clearly named panel-source folder.
Put deterministic final clip storyboard boards into
```
final_image_package/clip_storyboards/
```
.
Put character, scene, product, prop, expression, and pose references into
```
final_image_package/support_assets/
```
.

Keep filenames traceable:

<image-id>__zh__v01.png

<image-id>__en__v01.png

<image-id>__selected.png

If direct filesystem saving is unavailable from the image tool, still create prompt files and record intended output paths. Mark image generation as blocked instead of implying reference-conditioned images exist.

生成图像时：

将使用中文提示词生成的图像放入
```
05_images/zh/
```
。
将使用英文提示词生成的图像放入
```
05_images/en/
```
。
将选定的干净 SceneDance 输入放入
```
05_images/selected/
```
。
将最终干净的片段关键帧放入
```
final_image_package/clip_keyframes/
```
。
将生成的干净分镜画面板源文件放入
```
final_image_package/clip_storyboards/panels/
```
或另一个名称明确的 panel-source 文件夹。
将确定性的最终片段分镜板放入
```
final_image_package/clip_storyboards/
```
。
将角色、场景、产品、道具、表情和姿势参考图放入
```
final_image_package/support_assets/
```
。

保持文件名可追踪：

<image-id>__zh__v01.png

、

<image-id>__en__v01.png

、

<image-id>__selected.png

。

如果图像工具无法直接保存到文件系统，仍应创建提示词文件并记录预期输出路径。将图像生成标记为 blocked，不得暗示已存在基于参考条件生成的图像。

Storyboard Board Contract

分镜板契约

Final

clip_storyboards/

files are deterministic production boards:

Each board covers exactly one
```
CLIP###
```
.
Each board uses clean visual panels or keyframes as image inputs; the board layout and all readable text are rendered by local deterministic tooling.
Required visual panels:
```
START
```
/
```
KEY ACTION
```
/
```
EDIT OUT
```
. Add
```
RECEIVE IN
```
or
```
HANDOFF
```
only when it clarifies the boundary.
Required readable fields: project/title,
```
CLIP ID
```
, time range, duration, scene/location, aspect ratio, style/tone, camera method, action start, key action, edit-out state, emotion start/end, receiver-in, handoff-out, edit type, audio bridge, reference-image combination, risk/fallback.
Required source mapping: each visual panel must map back to a keyframe/panel path and each text field must come from
```
clip_plan.md
```
,
```
shot_cards.md
```
,
```
handoff_design_matrix.md
```
,
```
edit_boundary_matrix.md
```
, or
```
scenedance_shot_prompts.md
```
.
Do not leave blank text boxes or placeholder captions in the final board.
Do not rely on generated in-image text for production metadata. If an image model creates text, treat it as decorative noise and replace the board with a deterministic render.
Do not call a final board complete until it can be read by a human editor without opening the Markdown files.

最终

clip_storyboards/

文件必须是确定性的制作板：

每块分镜板必须且只能覆盖一个
```
CLIP###
```
。
每块分镜板都使用干净视觉画面板或关键帧作为图像输入；版面布局和所有可读文字均由本地确定性工具渲染。
必需视觉画面板：
```
START
```
/
```
KEY ACTION
```
/
```
EDIT OUT
```
。仅在有助于说明边界时添加
```
RECEIVE IN
```
或
```
HANDOFF
```
。
必需可读字段：项目/标题、
```
CLIP ID
```
、时间范围、时长、场景/地点、宽高比、风格/基调、镜头方法、动作开始、关键动作、剪出状态、情绪开始/结束、接收状态、交接输出、剪辑类型、音频桥接、参考图组合、风险/回退方案。
必需的来源映射：每个视觉画面板必须能够映射回关键帧/画面板路径；每个文字字段必须来自
```
clip_plan.md
```
、
```
shot_cards.md
```
、
```
handoff_design_matrix.md
```
、
```
edit_boundary_matrix.md
```
或
```
scenedance_shot_prompts.md
```
。
最终分镜板中不得留有空白文本框或占位字幕。
不得依赖生成图像中的文字承载制作元数据。如果图像模型生成了文字，应将其视为装饰噪声，并使用确定性渲染结果替换该分镜板。
在人工剪辑师无需打开 Markdown 文件就能读懂全部信息之前，不得将最终分镜板视为完成。

Validation Before Delivery

交付前验证

Before finalizing:

no SceneDance clip exceeds
```
15s
```
every new project shot defaults to
```
SH### = CLIP###
```
every shot card has all required YAML keys
every shot has action start/end and emotion start/end

every shot has

receiver_in

handoff_out

motion_vector

spatial_bridge

occlusion_carrier

visual_bridge

, and

handoff_risk_reduction

every shot has a primary clean input keyframe ID
every shot lists its reference image combination
every neighboring pair has a handoff design row
every neighboring pair has an edit boundary row
every boundary has a cut type, matching logic, audio bridge, CapCut handling, risk, and fallback, and references the handoff logic
every recurring character/product/scene uses bible IDs
Chinese and English Image 2 prompts are separated
storyboard boards are not treated as the only SceneDance video input
storyboard boards were assembled deterministically from clean panels/keyframes, not accepted as raw AI-generated board layouts
every final storyboard board contains readable
```
CLIP ID
```
, time range, start/key/edit-out structure, camera/action/edit metadata, handoff, edit boundary, audio bridge, reference-image combination, and risk/fallback
every final storyboard board has no blank caption areas, placeholder labels, missing timecode, mixed-clip layout, poster-like composition, or garbled production text
all promised images are present or explicitly marked blocked
final storyboard image count equals the final
```
CLIP###
```
count; overview/contact sheet images do not count
```
final_image_package/image_manifest.md
```
lists every delivered image and purpose

最终交付前：

不得有超过
```
15s
```
的 SceneDance 片段
每个新项目镜头默认采用
```
SH### = CLIP###
```
每张镜头卡都包含全部必需 YAML 键
每个镜头都有动作开始/结束和情绪开始/结束

每个镜头都有

receiver_in

、

handoff_out

、

motion_vector

、

spatial_bridge

、

occlusion_carrier

、

visual_bridge

和

handoff_risk_reduction

每个镜头都有主要的干净输入关键帧 ID
每个镜头都列出其参考图组合
每一对相邻镜头都有交接设计行
每一对相邻镜头都有剪辑边界行
每个边界都包含剪辑类型、匹配逻辑、音频桥接、CapCut 处理方式、风险和回退方案，并引用对应的交接逻辑
每个重复出现的角色/产品/场景都使用圣经 ID
Image 2 中文提示词和英文提示词相互分离
分镜板不得被视为唯一的 SceneDance 视频输入
分镜板必须由干净画面板/关键帧确定性拼装，不得直接接受原始 AI 生成版面
每张最终分镜板都包含可读的
```
CLIP ID
```
、时间范围、起始/关键动作/剪出结构、镜头/动作/剪辑元数据、交接、剪辑边界、音频桥接、参考图组合和风险/回退方案
每张最终分镜板都不得存在空白字幕区域、占位标签、缺失时间码、混合片段版面、海报式构图或乱码制作文字
所有承诺交付的图像均已存在，或已明确标记为 blocked
最终分镜图数量等于最终
```
CLIP###
```
数量；总览图/联系表不计入其中
```
final_image_package/image_manifest.md
```
列出了每张已交付图像及其用途