Seedance 2 Prompt Engineering Skill

# 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**
- `@asset` mapping (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

1. Always declare mode first.
2. Always include an explicit **Assets Mapping** section.
3. Use timecoded beats with one major action per segment.
4. Keep prompts concise and controllable (avoid vague poetic-only wording).
5. Add negative constraints when user needs clean output.
6. **Be specific and visual** — "a woman in a red trench coat walks through rain-soaked neon streets" >> "a woman walking".
7. **Separate dialogue/sound from visuals** — write dialogue with character name + emotion tag, then sound effects as a distinct layer.
8. **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)

1. **Mode**
2. **Assets Mapping**
3. **Final Prompt**
4. **Negative Constraints**
5. **Generation Settings**

Example skeleton:

```text
Mode: All-R...