Halloween DJ Dance Party · Open Skill 公开技能

Halloween DJ Dance Party

Halloween DJ Dance Party 是 Digen 上的公开 Open Skill。打开后即可了解该工作流能制作什么。

@eggs

Custom

1 0 1

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284862/2026-10-19/sc9teu74.jpg

# Halloween DJ Dance Party (Minimax H3)

Turn one uploaded character photo into a short vertical video of that same character dancing at a Halloween DJ party. The uploaded person is the hero of the clip: identity, face and hair must stay recognizable, only the outfit, set and lighting are re-staged into a spooky club.

This Skill has no structured project state. Progress lives in the conversation, in canvas assets, and in `write_todos`.

## When to use

- The user uploads a photo of a person and asks for a Halloween / DJ / dance-party video of them.
- The user uploads a photo and asks to change the character's outfit first, then animate it.
- The user asks for a "Halloween DJ dance" style clip, costume-swap dance clip, or party-loop clip from a reference image.
- The user wants a still-only costume preview of the character in Halloween DJ styling (image only; skip the video steps).

## Inputs and outputs

| Input | Required | How it is used |
|---|---|---|
| Reference image of the character | Yes | Identity source. Passed to `recognize_image` for identity locking, used as `input_urls` for the costume frame (`i2i.hd.lite`) and for the video (`ref2v`). Never replaced by a newly invented face. |
| Text | Optional | Supplies the outfit/costume, dance energy, mood, duration, or aspect ratio. Every concrete detail the user typed must appear in the costume plan and in the video prompt. |

Output: one short vertical video (`entity_type='final_videos'`), plus an optional costume preview still (`entity_type='characters'`) when the user wants to approve the look first.

## Steps

1. Read `references/workflow-stages.md` and `references/style-and-prompt.md` with `read_skill_file` before doing anything else. Use the stage table to decide which stage the user is in; use `references/style-and-prompt.md` for style, subject, layout, must-keep and must-avoid wording.
2. **Lock identity.** Call `recognize_image(image_url=<uploaded image>, prompt="Describe the person's face, hair, eyes, skin, makeup, body type and any distinctive marks in detail")`. Write the identity lock list (face shape, hair color and length, eye color, freckles/marks, makeup, body proportions, visible accessories). Never guess silently — if the image is missing, ask for it; if several people are in it, ask which one is the main character.
3. **Customize the outfit.** Merge the user's text with the default costume rules in `references/style-and-prompt.md`. State the costume plan to the user in 1–3 lines. If the user wants to see the look first, or the costume is complex, call `generate_image(model='i2i.hd.lite', input_urls=[<uploaded image>], prompt=<costume frame prompt from references/style-and-prompt.md>, orientation='portrait', name='<character name> - Halloween DJ look', entity_type='characters')` and use that frame as the video reference. If the user only wants this still, stop here.
4. **Write the video script before any video generation.** Call `read_skill_file(name='halloween-dj-dance-video', path='references/video-script.md')`, then write a complete English script for THIS request — logline, runtime, identity lock, wardrobe, set and decor, lighting, camera plan, beat sheet with time ranges, motion continuity, must keep, must avoid. Fill it with this user's costume words, mood and image analysis; a generic template is not acceptable.
5. **Build the video prompt.** Call `read_skill_file(name='halloween-dj-dance-video', path='references/video-content-constraints.md')` and compress the script into a single English video prompt using the template in `references/style-and-prompt.md` (order: identity + wardrobe → dance beats → set/decor → lighting → camera → quality). Keep it specific and 60–110 words.
6. **Generate the video.** Call `generate_video(model='ref2v', input_urls=[<costume frame if generated, otherwise the uploaded image>], prompt=<the prompt from step 5>, orientation='portrait', duration=5, name='<character name> - Halloween DJ Dance Party', entity_type='final_