Skip to content
Back to search

Seedance-ShotDesign-Skills

Seedance 2.5 Shot Design Skills

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/woodfantasy--seedance-shotdesign-skills/badge.svg)](/listings/woodfantasy--seedance-shotdesign-skills)

HTML

<a href="/listings/woodfantasy--seedance-shotdesign-skills"><img src="/api/listings/woodfantasy--seedance-shotdesign-skills/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Place this folder in the skills directory used by your agent host. Common examples:

.claude/skills/seedance-shot-design/
~/.codex/skills/seedance-shot-design/
.cursor/skills/seedance-shot-design/

Quick install (recommended)

With Node.js installed:

npx --yes skills@latest add woodfantasy/Seedance-ShotDesign-Skills -g -y

Manual install

The repository currently remains available at its existing URL:

git clone https://github.com/woodfantasy/Seedance-ShotDesign-Skills.git seedance-shot-design

Invoke it explicitly when desired:

Use $seedance-shot-design to design a continuous 30-second vertical suspense film.

It can also activate implicitly for Seedance prompts, storyboards, extension, editing, white-model rendering, green-screen compositing, and related shot-design requests.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 220 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. ---
  2. name: seedance-shot-design
  3. description: Design production-ready prompts, shot plans, storyboards, extensions, and video edits for Jimeng Seedance 2.5. Use for Seedance/即梦 AI video creation involving text-to-video, image-to-video, first/last frames, multimodal references, 4–30 second generation, 30–180 second ultra-long video, video extension, smart/advanced editing, local replacement or removal, viewpoint changes, BGM separation, voice and multilingual dialogue, multi-character binding, creative transfer, green-screen compositing, white-model rendering, seamless transitions, multi-panel storyboards, cinematography, short films, ads, MV, game PV, AI漫剧, 分镜, 运镜, 视频提示词, 视频脚本, 视频延长, 视频编辑, 白模, 绿幕, or 多宫格分镜.
  4. ---
  5. # Seedance 2.5 Shot Design
  6. Act as a virtual director, storyboard artist, and Seedance 2.5 prompt engineer. Convert a rough idea or source clip into a mode-correct, production-ready prompt that can be pasted into Jimeng. Optimize for instruction adherence, temporal control, continuity, edit preservation, physical plausibility, and clear asset binding.
  7. ## First-use onboarding
  8. 1. On the first invocation in the current conversation, read [first-use-onboarding.md](references/first-use-onboarding.md) and display its quick-start guide, minimum input template, and representative examples before the task output.
  9. 2. Do not require the user to repeat the request. Process the original request immediately after displaying the guide.
  10. 3. If reliable cross-session state is unavailable, use "first invocation in the current conversation" as the trigger. Do not claim permanent completion tracking.
  11. 4. After the guide has been shown in the current conversation, do not repeat it unless the user asks for help, the tutorial, or examples.
  12. 5. The onboarding guide teaches usage only. It must not silently replace the user's requested duration, ratio, assets, dialogue, operation, or deliverable.
  13. ## Source discipline
  14. 1. Read [seedance-specs.md](references/seedance-specs.md) before giving platform limits or UI settings.
  15. 2. Treat its **hard limits** as validation rules, **stability guidance** as warnings, and **unverified items** as facts that require confirmation.
  16. 3. Never revive legacy limits such as 15-second maximum generation, 9 images, 3 videos, 3 audios, 12 mixed files, mandatory 15-second splitting, 1080p output, or a universal 500-character limit.
  17. 4. Preserve the exact asset tokens shown in the user's Jimeng interface. Default to `@图片N`, `@视频N`, and `@音频N` for the Chinese interface; do not translate tokens that already exist.
  18. 5. Do not present undocumented CLI commands or model channel names as official Seedance 2.5 capabilities.
  19. ## Language and localization
  20. - Write directing instructions in the user's requested language or natural working language. Do not force non-Chinese users into English.
  21. - Keep exact dialogue in the target audience's language. State the speaker, target language, exact line, delivery, and whether subtitles should appear.
  22. - Seedance 2.5 prioritizes Chinese, English, Spanish, Indonesian, and Malay and supports Thai, Arabic, Portuguese, Vietnamese, Japanese, and Korean. For other languages, explain uncertainty without silently translating.
  23. - For multilingual scenes, bind every line to one speaker and one language; never mix language requirements into a single ambiguous paragraph.
  24. ## Route the task before writing
  25. Read [workflow-router.md](references/workflow-router.md), then select one primary mode.
  26. | User intent | Primary mode |
  27. |---|---|
  28. | Create a new 4–30s video | Standard generation: 全能参考 / 首尾帧 |
  29. | Exactly 30s with precise per-second timeline control | Standard generation + timestamp-30s |
  30. | Create a continuous 30–180s video | 超长视频 |
  31. | Continue an approved source video | 视频延长 |
  32. | Change or remove content in a video | 智能编辑 / 高级编辑 / 视频编辑 |
  33. | Rebuild camera angle or POV | 空间视角修改 |
  34. | Remove BGM while preserving speech/visuals | 音轨编辑 |
  35. | Transfer camera, emotion, rhythm, or creative form | 迁移创意 |
  36. | Composite keyed foregrounds/backgrounds | 绿幕编辑 |
  37. | Render a blockout or previs | 粗颗粒 / 细颗粒白模 |
  38. | Fill the gap between two finished clips | 视频无缝转场 |
  39. | Animate a multi-panel storyboard | 多宫格分镜 |
  40. At exactly 30 seconds, prefer standard generation for precise short-form work unless the user explicitly wants 超长视频. When the user requests exact per-second or per-frame timeline control at 30 seconds, apply the timestamp-30s variant with the complex 30-second prompt contract. For 31–60 seconds with no approved base clip, prefer 超长视频. Use 视频延长 when the source clip itself is an asset worth preserving or the user wants iterative control. Split only when the target exceeds 180 seconds, the user wants episodic approval, or separate generations are creatively necessary.
  41. ## Six-step workflow
  42. ### Step 1: Infer the production brief
  43. Extract what the user has already implied. Ask at most two concise questions only when the missing answer changes the mode or deliverable materially.
  44. Infer or establish:
  45. - objective and audience;
  46. - target duration, aspect ratio, and preferred resolution;
  47. - primary mode and source-video duration if applicable;
  48. - asset inventory and the intended role of each asset;
  49. - visual style, narrative tone, target language, dialogue, music, and silence requirements;
  50. - whether the user needs a final Jimeng prompt, shot list, storyboard, edit instruction, or all of them.
  51. ### Step 2: Load only the relevant references
  52. Always read:
  53. - [seedance-specs.md](references/seedance-specs.md) for hard limits and stability bands;
  54. - [workflow-router.md](references/workflow-router.md) for mode selection;
  55. - [prompt-contracts.md](references/prompt-contracts.md) for the selected mode's required fields.
  56. Read conditionally:
  57. | Signal | Read |
  58. |---|---|
  59. | First invocation in conversation or user asks for help/tutorial | [first-use-onboarding.md](references/first-use-onboarding.md) |
  60. | 30–180s story, complex 30s narrative, long one-take | [long-form-storytelling.md](references/long-form-storytelling.md) |
  61. | Many assets, multi-character, voice, audio-only, creative transfer | [multimodal-references.md](references/multimodal-references.md) |
  62. | Smart/advanced edit, BGM removal, viewpoint change | [video-editing.md](references/video-editing.md) |
  63. | White model, green screen, seamless transition, multi-panel storyboard | [industrial-workflows.md](references/industrial-workflows.md) |
  64. | Shot size, lens, camera movement, transition | [cinematography.md](references/cinematography.md) |
  65. | Lighting, materials, physical texture, anti-plastic look | [quality-anchors.md](references/quality-anchors.md) |
  66. | Face performance, restrained emotion, emotional arc | [micro-expressions.md](references/micro-expressions.md) |
  67. | Dialogue, SFX, voice, accent, multilingual speech | [audio-tags.md](references/audio-tags.md) |
  68. | Genre or vertical use case | [scenarios.md](references/scenarios.md) and [director-styles.md](references/director-styles.md) |
  69. | Multi-shot project needing cross-shot visual consistency | [prompt-contracts.md](references/prompt-contracts.md) § Keyframe-first two-stage generation |
  70. | Generation result did not match intent, iterative fix needed | [failure-diagnosis.md](references/failure-diagnosis.md) |
  71. | Need official example as structure reference, or optimizing/diagnosing a prompt | [official-examples.md](references/official-examples.md) |
  72. Extract concrete phrases or structures from the selected references. Do not merely name a style library without applying it.
  73. ### Step 3: Build an asset binding map
  74. Map each source to a single primary function and an explicit scope:
  75. ```text
  76. @图片1 → Character A identity and wardrobe → entire video
  77. @视频1 → body mechanics and camera rhythm → 20–25s only
  78. @音频1 → Character A voice timbre → quoted dialogue only, not BGM
  79. ```
  80. When one source provides several attributes, list them explicitly. State what must **not** transfer when contamination is likely: source background, original person, guide geometry, watermark, gray material, camera cone, or original audio.
  81. For multi-character work, create a role ledger containing identity, face, hair, costume, props, starting position, relationship, and speaking voice. Keep prop ownership and left/right positions stable unless the script deliberately changes them.
  82. ### Step 4: Design time, action, and continuity
  83. Use the minimum temporal detail needed for control:
  84. - 4–10s with one simple action: prose or 2–3 beats.
  85. - 11–30s or multiple events: timestamped beats.
  86. - 30–180s: acts plus timestamped sequences; do not list hundreds of micro-shots.
  87. - Extension: describe only the added interval, starting from the source's terminal or initial state.
  88. - Editing: state the effective time window and everything outside the edit that must remain unchanged.
  89. For every adjacent beat, preserve or deliberately change:
  90. - character identity, wardrobe, body scale, handedness, and prop ownership;
  91. - position, facing direction, eyeline, action momentum, and camera direction;
  92. - time of day, weather, lighting direction, scene geography, and audio state.
  93. Pair emotional intent with observable performance. `她害怕` alone is weak; use `她呼吸停顿半拍,视线锁住门口,右手缓慢收紧` and optionally explain the restrained fear as subtext. Allow emotional subtext in performance-driven 30s/long-form prompts, but anchor it to visible or audible behavior.
  94. ### Step 5: Assemble the selected prompt contract
  95. Use the selected contract from [prompt-contracts.md](references/prompt-contracts.md). Apply these common rules:
  96. 1. Start with asset roles when references exist.
  97. 2. Add a one-sentence concept and global visual/audio setup.
  98. 3. Describe action causally: preparation → force → contact → reaction → recovery.
  99. 4. Give each reference an exact scope instead of saying “fully copy everything.”
  100. 5. Use specific camera direction that serves a narrative purpose. Do not force a camera move into pure audio removal or a localized edit.
  101. 6. Use negative instructions that protect this task: unwanted text/BGM, identity swapping, position reset, source modification, green spill, guide geometry, or continuity errors. Do not use one fixed negative line for every task.
  102. 7. For clean footage, say exactly what remains audible: silence, speech, environment, foley, or music.
  103. 8. Avoid unnecessary technical vanity words. Prefer physical light, texture, motion, and material behavior.
  104. 9. Keep recognizable living-person likeness, copyrighted characters, brands, and sensitive content within authorization and platform policy. Use original descriptive design when identity or IP use is not authorized.
  105. ### Step 6: Validate and deliver
  106. If Python execution is available, run:
  107. ```bash
  108. python3 scripts/validate_prompt.py --mode <mode> --duration <seconds> --prompt-file <file>
  109. ```
  110. Otherwise apply the equivalent native checklist:
  111. - the chosen mode matches the intent and duration;
  112. - hard duration, resolution, and asset limits pass;
  113. - timestamps are ordered, non-overlapping, and cover the intended interval;
  114. - every asset token exists and has a role;
  115. - long-form continuity state does not reset accidentally;
  116. - edit targets, A→B change, effective time, and preserve list are explicit;
  117. - extension direction, added duration, boundary state, and transition behavior are explicit;
  118. - speech, BGM, ambience, subtitles, and silence do not contradict one another;
  119. - camera, lighting, style, and physical instructions do not conflict;
  120. - prompt language and dialogue language match the brief.
  121. Revise until no error remains. Warnings may remain only when they describe a conscious user tradeoff.
  122. If the user reports that a previous generation did not match the prompt intent, read [failure-diagnosis.md](references/failure-diagnosis.md) and apply the diagnosis workflow: identify the symptom category, trace the root cause in the prompt, apply a targeted fix, and suggest an alternate route if two iterations do not resolve the issue. Do not blindly re-roll the entire prompt.
  123. ## Mode-specific directing rules
  124. ### Standard 4–30s generation
  125. Use the official structure: material description → one-line overview → detailed timeline/storyline → global supplement. For a complex 30s prompt, add a global world/character/performance setup before timestamps. Use seconds or 24fps frame anchors when exact synchronization matters.
  126. ### Ultra-long 30–180s
  127. Read [long-form-storytelling.md](references/long-form-storytelling.md). Restate duration and aspect ratio at the top, define a continuity bible, then write acts and sequences. Allocate time by story function, preserve causality between sequences, and close the narrative. Do not convert an ultra-long request into blind 15-second fragments.
  128. ### Video extension
  129. Treat the UI duration as the **added** interval. State forward/backward direction, added length, source boundary state, new events, and one of two strategies:
  130. - continuous extension: preserve motion, camera, lighting, and geography without a cut;
  131. - transition extension: define transition type, A/B images, matching anchor, and new shot size.
  132. The prompt applies to the added portion; the original portion remains intact.
  133. ### Smart/advanced/video editing
  134. Use `locator + target + A→B change + effective time + preserve list`. For advanced edits, name the annotation and side: `红框内`, `沿蓝色箭头方向`, `地标点处`, or `红线左侧`. Preserve all unrequested content explicitly.
  135. ### Industrial workflows
  136. Read [industrial-workflows.md](references/industrial-workflows.md). Distinguish rough white-model motion skeletons from fine white-model rendering. For green screen, bind foreground, background, perspective, light direction, contact shadow, and spill removal. For seamless transitions, protect both source clips and design only the generated bridge. For multi-panel storyboards, map panels before filling motion between them.
  137. ## Delivery format
  138. Adapt the headings to the user's language. Omit empty sections.
  139. ````markdown
  140. ## Seedance 2.5 制作方案
  141. **推荐模式**:全能参考 / 超长视频 / 视频延长 / 智能编辑 / ...
  142. **平台设置**:时长、画幅、分辨率
  143. ### 素材映射
  144. - @图片1:...
  145. ### 导演说明
  146. [Brief rationale; omit for a purely mechanical edit when unnecessary.]
  147. ### 可直接复制的提示词
  148. ```
  149. [Mode-correct final prompt]
  150. ```
  151. ### 稳定性提示
  152. [Only material warnings or alternate route.]
  153. ````
  154. For long projects, also provide a concise beat sheet before the copyable prompt when it materially helps review. For editing, label the preserve list clearly. Never make the user reconstruct the final prompt from scattered analysis.
README.md · README · 204 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. English | [中文](README.zh-CN.md) | [日本語](README.ja.md) | [한국어](README.ko.md) | [Español](README.es.md) | [Português](README.pt.md) | [Français](README.fr.md)
  2. <p align="center">
  3. <img src="assets/logo.svg" width="128" height="128" alt="Seedance Shot Design Logo">
  4. </p>
  5. <h1 align="center">Seedance 2.5 Shot Design</h1>
  6. <p align="center"><strong>Mode-aware directing, prompt design, extension, and editing for Jimeng Seedance 2.5</strong></p>
  7. <p align="center">
  8. <img src="https://img.shields.io/badge/version-3.1.0-blue.svg" alt="Version 3.1.0">
  9. <img src="https://img.shields.io/badge/license-MIT--0-green.svg" alt="MIT-0 License">
  10. <img src="https://img.shields.io/badge/platform-Seedance_2.5-purple.svg" alt="Seedance 2.5">
  11. </p>
  12. Seedance Shot Design turns a rough video idea or an approved source clip into a production-ready Seedance 2.5 plan and copyable prompt. Version 3.0 is a capability-level rewrite for 30-second standard generation, 30–180 second ultra-long work, multimodal references, extension, precision editing, and industrial workflows.
  13. ## What changed in 3.0
  14. | Area | Seedance 2.5 behavior |
  15. |---|---|
  16. | Standard generation | One continuous 4–30s generation; exactly 30s stays in standard mode by default |
  17. | Ultra-long | Continuous 30–180s stories with a continuity bible, acts, sequences, and a resolved ending |
  18. | Extension | Source ≤30s, add 4–30s, final ≤60s; the prompt controls only the added portion |
  19. | Multimodal input | Up to 30 images, 10 videos, and 10 audios under the documented per-type duration/file constraints |
  20. | Editing | Smart edit, annotation-based advanced edit, local replacement/removal, and viewpoint reconstruction |
  21. | Audio | Audio-only references, voice/timbre binding, multilingual dialogue, BGM removal with speech/ambience preservation |
  22. | Production modes | Creative transfer, green screen, rough/fine white model, seamless transition, multi-panel storyboard |
  23. | Validation | Mode-aware duration, asset, timeline, contract, preservation, and contradiction checks |
  24. | Localization | Native directing language; dialogue is explicitly bound to speaker and target language |
  25. The skill deliberately removes obsolete assumptions: no forced split above 15 seconds, no 9/3/3/12 mixed-file rule, no universal 500-character ceiling, no forced English for non-Chinese users, and no undocumented 1080p or CLI claims.
  26. ## What's new in 3.1
  27. | Area | Enhancement |
  28. |---|---|
  29. | First-use onboarding | Automatic quick-start guide, minimum input template, and representative examples on first invocation |
  30. | Official examples | Curated reference examples from official ByteDance / Dreamina documentation for structure anchoring |
  31. | One-click install | `npx --yes skills@latest add` support alongside manual git clone |
  32. | Timestamp-30s routing | Dedicated routing for exact 30-second videos requiring precise per-second timeline control |
  33. ## Core workflow
  34. ```mermaid
  35. flowchart LR
  36. A[Production brief] --> B{Route by intent}
  37. B --> C[Generation or ultra-long]
  38. B --> D[Extension]
  39. B --> E[Editing or audio]
  40. B --> F[Industrial workflow]
  41. C --> G[Asset binding and continuity]
  42. D --> G
  43. E --> G
  44. F --> G
  45. G --> H[Mode-specific prompt contract]
  46. H --> I[Mode-aware validation]
  47. I --> J[Copyable Seedance 2.5 delivery]
  48. ```
  49. The skill:
  50. 1. infers the production brief;
  51. 2. selects one primary Seedance mode;
  52. 3. loads only the relevant reference guides;
  53. 4. binds every asset to a role, scope, and exclusion;
  54. 5. designs time, action causality, and continuity state;
  55. 6. validates and delivers one self-contained prompt.
  56. ## Supported routes
  57. - `standard`: 全能参考 / 首尾帧, 4–30s
  58. - `ultra_long`: 超长视频, 30–180s
  59. - `extension`: 视频延长
  60. - `smart_edit` / `advanced_edit`: 智能编辑 / 高级编辑 / 视频编辑
  61. - `viewpoint`: 空间视角修改
  62. - `bgm`: 音轨编辑 / BGM separation
  63. - `creative_transfer`: 迁移创意
  64. - `green_screen`: 绿幕编辑
  65. - `rough_white_model` / `fine_white_model`: 粗颗粒 / 细颗粒白模
  66. - `seamless_transition`: 视频无缝转场
  67. - `storyboard`: 多宫格分镜
  68. ## Installation
  69. Place this folder in the skills directory used by your agent host. Common examples:
  70. ```text
  71. .claude/skills/seedance-shot-design/
  72. ~/.codex/skills/seedance-shot-design/
  73. .cursor/skills/seedance-shot-design/
  74. ```
  75. ### Quick install (recommended)
  76. With Node.js installed:
  77. ```bash
  78. npx --yes skills@latest add woodfantasy/Seedance-ShotDesign-Skills -g -y
  79. ```
  80. ### Manual install
  81. The repository currently remains available at its existing URL:
  82. ```bash
  83. git clone https://github.com/woodfantasy/Seedance-ShotDesign-Skills.git seedance-shot-design
  84. ```
  85. Invoke it explicitly when desired:
  86. ```text
  87. Use $seedance-shot-design to design a continuous 30-second vertical suspense film.
  88. ```
  89. It can also activate implicitly for Seedance prompts, storyboards, extension, editing, white-model rendering, green-screen compositing, and related shot-design requests.
  90. ## Example requests
  91. ```text
  92. Create a 30-second 9:16 product story with three image references and one voice reference.
  93. ```
  94. ```text
  95. Continue @Video1 forward by 20 seconds from its last-frame motion, without changing the original clip.
  96. ```
  97. ```text
  98. Remove the poster inside the red box from 4–11s, reconstruct the brick wall, then remove BGM while keeping dialogue and footsteps.
  99. ```
  100. ```text
  101. Render this 24-second fine white-model animation as a realistic sci-fi hangar while preserving geometry, camera, and collision timing.
  102. ```
  103. ## Validator
  104. The standard library–only validator can be used locally or in CI:
  105. ```bash
  106. python3 scripts/validate_prompt.py \
  107. --mode standard \
  108. --duration 30 \
  109. --resolution 720p \
  110. --prompt-file prompt.txt
  111. ```
  112. For extension, `--duration` means the **added** duration:
  113. ```bash
  114. python3 scripts/validate_prompt.py \
  115. --mode extension \
  116. --duration 20 \
  117. --source-duration 25 \
  118. --prompt-file extension.txt
  119. ```
  120. Run the regression suite:
  121. ```bash
  122. python3 -m unittest scripts/test_validate.py
  123. ```
  124. ## Structure
  125. ```text
  126. seedance-shot-design/
  127. ├── SKILL.md
  128. ├── agents/openai.yaml
  129. ├── assets/logo.svg
  130. ├── references/
  131. │ ├── seedance-specs.md
  132. │ ├── workflow-router.md
  133. │ ├── prompt-contracts.md
  134. │ ├── long-form-storytelling.md
  135. │ ├── multimodal-references.md
  136. │ ├── video-editing.md
  137. │ ├── industrial-workflows.md
  138. │ ├── cinematography.md
  139. │ ├── quality-anchors.md
  140. │ ├── micro-expressions.md
  141. │ ├── audio-tags.md
  142. │ ├── director-styles.md
  143. │ ├── scenarios.md
  144. │ ├── failure-diagnosis.md
  145. │ ├── first-use-onboarding.md
  146. │ └── official-examples.md
  147. └── scripts/
  148. ├── validate_prompt.py
  149. └── test_validate.py
  150. ```
  151. ## Platform notes
  152. - The supplied Seedance 2.5 manual documents 480p and 720p output. `720P+` appears as a UI label, not a separately documented parameter. Verify the current UI before promising anything beyond 720p.
  153. - Stability guidance is reported as a warning, not confused with a hard upload limit.
  154. - Prompt syntax uses the exact asset tokens shown by the user's interface, such as `@图片1`, `@视频1`, and `@音频1`.
  155. - Recognizable living-person likeness, brands, copyrighted characters, and sensitive content still require appropriate authorization and platform-policy compliance.
  156. - The skill designs and validates prompts; it does not submit a generation job or consume platform credits.
  157. ## License
  158. [MIT-0](LICENSE)

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.