Chinese Classical Dance Video · публичный навык Open Skill

Chinese Classical Dance Video

Chinese Classical Dance Video — публичный Open Skill на Digen. Откройте, чтобы увидеть, что создаёт этот процесс.

@JackRhino

Trailer

0 0 0

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284865/2026-10-22/kkvbyqdq.jpg

# Chinese Classical Dance Video

Turn a user-supplied photo into a short, graceful Chinese classical dance clip. The uploaded subject (person, costume, face, hair, ornaments) is preserved and re-staged inside one consistent visual world: peach-blossom petals, translucent silk ribbons, warm backlight, soft mist. Image in, moving dance clip out.

There is no structured project state. Progress lives in the conversation, the canvas assets, and `write_todos`.

## When to use

- The user uploads a portrait/photo and asks for a dance video, a "make her dance" clip, an animated version, or a moving poster.
- The user supplies a first frame and a last frame and wants a smooth dance transition between them.
- The user supplies text only (a description of a dancer) and wants an image-driven dance clip.
- The user wants an existing dance clip redone: new pose, new camera move, longer/shorter, different mood, different costume color.
- Do not use for still-image-only requests; if the user only wants a picture, generate the image and stop.

## Steps

1. **Read the request and inventory the inputs.**
   - Call `write_todos` with the stages you will run (analysis, keyframe, script, video, review).
   - Count reference images: two (first + last frame), one (single still), or none (text only).
   - Note the user's text: mood, duration, camera preference, costume color, setting, any "keep this" instruction.

2. **Analyze every uploaded image.**
   - Call `recognize_image` on each reference image with a prompt asking for: subject identity cues (face shape, hairstyle, hair color, eye color, skin tone), costume and ornament details, dominant colors, framing, background elements, and light direction.
   - Record the findings in your reply as a short list. These are the must-keep items later.

3. **Read the visual rules before writing any prompt.**
   - Call `read_skill_file(name="chinese-classical-dance-video", path="references/style-and-prompt.md")`.
   - Follow its Style / Subject / Layout / Must keep / Must avoid sections and use its prompt templates. Do not invent a different visual language.

4. **Read the staged workflow and branch on the input type.**
   - Call `read_skill_file(name="chinese-classical-dance-video", path="references/workflow-stages.md")`.
   - It defines the branches: two images → straight to first/last-frame video; one image → derive a matching end frame first; text only → build a first frame from scratch. Follow the listed checks, fallbacks and naming conventions there, including `name` and `entity_type`.

5. **Prepare keyframes when the workflow asks for them.**
   - For a missing or weak end frame, or for a text-only request, call `generate_image` with the prompt assembled from `references/style-and-prompt.md`.
   - Use `model='i2i.hd.lite'` when transforming an uploaded image, and `model='t2i'` only when there is no source image at all.
   - Keep `orientation='portrait'` unless the user explicitly wants landscape. Pass a short `name` (for example `dance-keyframe-01`) and `entity_type='storyboard_images'`.
   - The end frame must stay the same character and costume as the start frame; only pose, ribbon flow and petal positions change.

6. **Read the video constraints.**
   - Call `read_skill_file(name="chinese-classical-dance-video", path="references/video-content-constraints.md")`.
   - It defines shot framing, camera movement, motion speed, ribbon and hair physics, duration feel, first/last-frame continuity limits, and everything that must never appear.

7. **Write the shot script for this exact request.**
   - Call `read_skill_file(name="chinese-classical-dance-video", path="references/video-script.md")`.
   - Fill its structure with the current user intent and the `recognize_image` findings: logline, duration, 4–6 beats, framing and camera per beat, subject action, light and atmosphere, transitions, must-keep, must-avoid.
   - This script is for this one request. Never reuse a generic template as the finished script.