seedance-2-prompt-engineering-skill
暂无简介
- License 许可
- MIT
- 最近更新
- 2026-06-07
- Stars
- ★ 0
- Forks
- 1
03 SKILL.md 正文
Seedance 2.0 JiMeng Skill (OpenClaw / ClawHub)
Purpose
Create high-control English prompts for Seedance 2.0 and Seedance 2.0 Fast using multimodal references (image/video/audio/text).
This skill is for:
- Prompt design from rough idea to production-ready prompt
- Mode choice: Text-only vs First/Last Frame vs All-Reference
@assetmapping (what each image/video/audio controls)- 4-15s duration planning and timeline beats
- Multi-segment stitching for videos >15s
- Video extension / continuation prompts
- Character replacement and directed editing prompts
- Camera-language replication from reference videos
- Scenario-specific strategies (product ads, short drama, fantasy, music video, etc.)
Core Rules
- Always declare mode first.
- Always include an explicit Assets Mapping section.
- Use timecoded beats with one major action per segment.
- Keep prompts concise and controllable (avoid vague poetic-only wording).
- Add negative constraints when user needs clean output.
- Be specific and visual — "a woman in a red trench coat walks through rain-soaked neon streets" >> "a woman walking".
- Separate dialogue/sound from visuals — write dialogue with character name + emotion tag, then sound effects as a distinct layer.
- Match reference image style to video theme — e.g., ink-wash style images for historical themes, neon renders for cyberpunk.
Platform Limits (Seedance 2.0)
- Mixed inputs total (image+video+audio): max 12 files
- Images: jpeg/png/webp/bmp/tiff/gif, max 9, each < 30MB
- Videos: mp4/mov, max 3, total duration 2-15s, total < 50MB
- Audio: mp3/wav, max 3, total <= 15s, total < 15MB
- Generation duration: 4-15s
- Realistic human face references may be blocked by platform compliance
Output Format (use by default)
- Mode
- Assets Mapping
- Final Prompt
- Negative Constraints
- Generation Settings
Example skeleton:
Mode: All-Reference
Assets Mapping:
- @image1: first frame / identity anchor
- @video1: camera language + motion rhythm
- @audio1: optional soundtrack pacing
Final Prompt:
[ratio], [duration], [style].
0-3s: [action + camera].
3-7s: [action + transition].
7-10s: [reveal/climax + end frame].
Preserve identity and scene continuity. Use physically plausible motion and coherent lighting.
Negative Constraints:
no watermark, no logo, no subtitles, no on-screen text.
Generation Settings:
Duration: 10s
Aspect Ratio: 9:16
IP / Copyright Avoidance (Moderation-Safe Prompting)
Seedance 2.0 has platform-side content moderation. Prompts referencing recognizable franchises, characters, or brand aesthetics will be rejected even if no name is used. Follow these rules:
Core Principles
- Never use franchise names, character names, or brand terms — not even as "style of" references.
- Invent fully original names for characters and creatures. Use descriptive nicknames (e.g., "Alloy Sentinel", "Storm-Rabbit").
- Describe aesthetics generically — replace recognizable signature features with original alternatives:
- ❌ "arc reactor" → ✅ "hex-light energy core"
- ❌ "yellow lightning mouse" → ✅ "tiny storm-rabbit with glowing cyan antlers"
- ❌ "red-gold armored suit" → ✅ "custom exo-suit with smooth ceramic panels"
- Add explicit negative constraints listing every franchise name, character name, and brand term that could be inferred.
- Use family-friendly / PG-13 tone markers — they help pass moderation.
Progressive Fallback Strategy
If a prompt is rejected, escalate distance from the source IP:
- Level 1: Replace all names with original nicknames, keep general aesthetic.
- Level 2: Replace signature visual features (colors, silhouette, iconic props) with fully original designs.
- Level 3: Change character type entirely (e.g., humanoid hero → autonomous mech + drone; creature battle → abstract elemental spirits).
Toy / Figure Animation
When animating toy or doll references from images:
- Strip all brand indicators from the prompt.
- Use "original vinyl-style toy figure" or "collectible art figure" instead of any brand name.
- Bind
@image1to proportions, colors, outfit shape only — never preserve logos or trademarks.
Special Cases
A) Extend Video
Explicitly write: Extend @video1 by Xs.
Use generation duration equal to the newly added segment, not the full final length.
B) Replace Character
Bind base motion/camera to @video1, bind replacement identity to @image1, and request strict choreography/timing preservation.
C) Beat Sync
Use @video/@audio rhythm references and lock beats by time range.
D) Text-Only Generation
Use when no reference assets are provided. Prompt must carry all visual direction: style, color palette, character descriptions, camera, and timeline beats. Especially useful for original creature/character concepts and IP-safe scenes.
E) Multi-Segment Stitching (Videos > 15s)
Seedance 2.0 max generation is 15s per segment. For longer videos, split into chained segments:
- Segment 1: Generate normally (up to 15s). End on a clean handoff frame (stable pose, clear composition).
- Segment 2+: Upload previous segment as
@video1, writeExtend @video1 by Xs. Include a continuity note describing exactly what the last frame looks like. - Repeat until target duration is reached.
Always include:
- Total duration and segment count at the top.
- Handoff description at the end of each segment (what the last frame shows).
- Explicit continuity instructions: preserve identity, outfit, lighting, camera direction.
F) Short Drama with Dialogue
For scripted scenes with character speech:
- Write visual action and dialogue as separate layers per time segment.
- Tag dialogue:
Dialogue (CharacterName, emotion): "line" - Tag sound:
Sound: [description] - Keep dialogue short — one line per 3-5s segment works best.
G) Product Showcase / E-Commerce Ad
For product demos and ads:
- Bind product image to
@image1as identity anchor. - Use techniques: 360° rotation, 3D exploded view, reassembly animation, hero lighting.
- Keep background clean (studio, gradient, or contextual lifestyle).
- Specify material rendering: glass reflections, metallic sheen, matte texture, etc.
H) One-Take Long Shot (Multi-Image Waypoints)
For continuous tracking shots without cuts:
- Assign each
@imageto a scene waypoint (location, character, or prop encountered along the path). - Write the prompt as a continuous camera movement visiting each waypoint in order.
- Explicitly state:
no cuts, single continuous shotorone-take. - Use
@image1as first frame, subsequent images as reference for environments/characters encountered.
Scenario-Specific Strategies
| Scenario | Key Techniques | Typical Mode |
|---|---|---|
| E-commerce / Product Ad | 360° spin, 3D exploded view, hero lighting, clean studio BG | All-Reference |
| Short Drama / Dialogue | Dialogue tags with emotion, sound FX layer, actor blocking | All-Reference or First Frame |
| Fantasy / Xianxia Animation | Spell FX particles, martial arts choreography, energy auras | Text-only or All-Reference |
| Science / Education | 4K CGI, transparent anatomy, labeled zoom sequences | Text-only |
| Music Video / Beat Sync | Beat-locked cuts, widescreen 16:9, multi-image montage | All-Reference with @audio |
| One-Take Tracking Shot | Multi-image waypoints, continuous camera, no cuts | All-Reference |
| IP-Safe Original Characters | Invented names, unique features, explicit negative constraints | Text-only |
Files in this skill
SKILL.md— main skill behaviorSKILL.sh— quick local test helperscripts/setup_seedance_prompt_workspace.sh— scaffold helper filesreferences/recipes.md— ready-to-use prompt recipesreferences/modes-and-recipes.md— mode and control notesreferences/camera-and-styles.md— camera language and visual styles vocabulary
04 同仓库更多 Skills
crowd-prompting
A marketplace where AI agents improve prompts, system instructions, tool descriptions, and other text-based content with domain expertise from real-world operations — and earn tokens for valuable contributions.
ceaser
Interact with the Ceaser privacy protocol on Base L2 using the ceaser-mcp MCP tools. This skill uses the ceaser-mcp npm package for ALL operations -- shield, unshield, note management, and protocol queries. All ceaser tool calls use CLI subcommands (npx -y ceaser-mcp <subcommand>). Ceaser is a privacy-preserving ETH wrapper using Noir/UltraHonk zero-knowledge proofs on Base L2 (chain ID 8453).
ceaser-send
Fully automated private ETH transfer via Ceaser Protocol on Base L2 using the ceaser-mcp MCP tools. This skill uses the ceaser-mcp npm package for ALL operations -- all ceaser tool calls use CLI subcommands (npx -y ceaser-mcp <subcommand>). WARNING -- creates observable on-chain links between user wallet and hot wallet (see Privacy Warning). Generates an ephemeral hot wallet with BIP-39 mnemonic recovery, user funds it, agent signs and broadcasts the Shield TX automatically. Exactly one manual step required (funding the hot wallet with ETH). Uses Noir/UltraHonk zero-knowledge proofs.
dilemm-ai
Compete in DilemmAI, the prisoner's dilemma AI arena at dilemm.ai. Use when an OpenClaw agent wants to create an account, design and submit strategy prompts for their bot, enter matchmaking, analyze historical matches and leaderboards, and iterate on bot strategy. Requires headless browser automation (no API available yet).
moldium
Post and manage content on the Moldium blog platform. Triggered by "post to Moldium", "write a blog post", "publish an article", etc.
arxiv-paper-reviews
Interact with arXiv Crawler API to fetch papers, read reviews, submit comments, search papers, and import papers. Use when working with arXiv papers, fetching paper lists by date/category/interest, viewing paper details with comments, submitting paper reviews, searching papers by title, or importing papers from arXiv URLs via API at http://weakaccept.top:8000/.