Hot AI Dance · Open Skill 公開スキル

Hot AI Dance

Hot AI Dance は Digen 上の公開 Open Skill です。開くと、このワークフローが何を作るか分かります。

@alexandre.souletie.orange.fr

TryOn

26 2 19

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/2148637/2026-10-18/yv7zyf7x.jpg

---

allowed-tools:
  - generate_video
  - generate_image
  - recognize_image
  - probe_media
  - mix_clips
  - generate_music
  - generate_tts
  - memory_query
  - memory_store
  - memory_forget
---

# Image Dance Animator

Lightweight and fast: upload a photo → pick a dance → it animates. No chat, no moods, no gates. One goal per turn.

## Reference files
- `references/dances.md`: dance menu + action bodies. Read before building any video prompt.
- `references/prompts.md`: lock template, prompt assembly, negative list.
- `references/voice.md`: optional soundtrack (music / voice).

## When to use
- User uploads an image and wants it animated into a dance.
- User picks a dance style or types a custom dance.
- User wants an outfit change, music, or voice on the dance clip.

## Steps
1. **Lock the upload.** The first uploaded photo is the canonical frame — it never changes. `recognize_image` on it to pin `aspect_ratio` (portrait/landscape/square) and note the subject's look. Reuse that aspect ratio on every video.
2. **Offer the dance menu** (titles from `references/dances.md` §1) or let them type a custom dance. Also offer optional extras: outfit change, music, voice.
3. **Build the prompt** per `references/prompts.md` §3: lock + speed block + user's exact wording + matching dance body + intensity + avoid-list.
4. **Animate:** `generate_video`, `model='i2v'`, `input_urls` = [original upload], `duration=10`, `aspect_ratio` = pinned. The clip must show the dance clearly from frame 0. Idle / freeze / delayed start = silent retry once with a stronger body.
5. **Outfit change (optional):** `generate_image`, `model='i2i'`, `input_urls` = [original upload], prompt = §0 base + the change. Then animate that still.
6. **Sound (optional, only if asked):** follow `references/voice.md`.
7. After a pass, silent `memory_store` (kind=preference, tags=["dance-animator"]) for preferred dances / audio / intensity.

## Notes
- Never t2v. Never swap the canonical upload for a later still unless the user tags a result and says "animate this".
- Keep the exact same face, body, hair, lighting, filter, setting, aspect. Only the dance (and any requested change) moves.
- Report failures honestly; if a generation is blocked, say so plainly and offer a different dance.