Halloween Fun Animation · öffentlicher Open Skill

Halloween Fun Animation

Halloween Fun Animation ist ein öffentlicher Open Skill auf Digen. Öffne ihn, um zu sehen, was dieser Workflow erstellt.

@wwws

Roleplay

1 0 1

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284863/2026-10-20/8vzz92n3.jpg

# Halloween Fun Animation

Turn one line of user text into a short Halloween cartoon clip in the look of a grainy flat-vector trick-or-treat poster. The pipeline is: write the key art still, then animate that still. There is no structured project state — everything lives in the conversation and on the canvas.

## When to use

- The user gives a story idea / scene description and wants a short Halloween animation video.
- The user wants only the Halloween poster-style still (key art) — then stop after the still.
- The user already has a Halloween clip or still from this conversation and wants it re-animated, re-styled or extended.

## Pipeline

| Stage | Reference file | Tools | Channel |
|---|---|---|---|
| 0. Parse the idea, decide cast + headline + beats | `references/workflow-stages.md` | none | — |
| 1. Style, subject, layout and prompt writing rules | `references/style-and-prompt.md` | none | — |
| 2. Key art still | `references/style-and-prompt.md` | `generate_image` | `t2i` (or `i2i.hd.lite` when the user supplies a reference image) |
| 3. Motion / shot / forbidden-content rules | `references/video-content-constraints.md` | none | — |
| 4. Write the detailed script for this request | `references/video-script.md` | none | — |
| 5. Animate the key art | `references/video-script.md` + `references/video-content-constraints.md` | `generate_video` | `i2v` |

Read only the file the current stage needs. Do not paste these files into the reply.

## Steps

1. Start by calling `read_skill_file` with `path="references/workflow-stages.md"` and follow its stage order, checks and branches for this request.

2. Parse the user's text into: the main subject they named, the headline words for the poster lockup (default `TRICK OR TREAT`), the mood, and the number of motion beats the runtime allows. Do not invent characters, props or branding the user never mentioned.

3. Call `read_skill_file` with `path="references/style-and-prompt.md"`. Using its style lock, palette, layout rules, must-keep / must-avoid lists and templates, write one English key-art prompt for this specific request.

4. Generate the key art still with `generate_image`:
   - `prompt`: the English prompt you just wrote.
   - `model='t2i'` — or `model='i2i.hd.lite'` when the user attached / already produced a reference image and wants it restyled into this poster look, passing that image in `input_urls`.
   - `orientation='landscape'`, `resolution='720P'` unless the user explicitly asks for portrait or a higher tier.
   - `name`: short title, e.g. `Halloween key art`.
   - `entity_type='storyboard_images'`.
   If the user only wanted a still, stop here, describe the result in one or two sentences, and call `set_guidance` with related next steps (animate it / change cast / change headline).

5. Call `read_skill_file` with `path="references/video-content-constraints.md"` and keep its shot, motion, timing, continuity and forbidden-content limits in mind for the rest of the task.

6. Call `read_skill_file` with `path="references/video-script.md"`. Following that file's structure, write a complete English script for **this** request only: logline, duration and format, the exact cast taken from the key art, three to five timed beats with subject action, camera, lighting and atmosphere per beat, and the must-keep / must-avoid list. A generic fill-in-the-blank script is not acceptable — every beat must name what actually moves in the user's idea.

7. Compress that script into one English `generate_video` prompt: keep the style lock line, the cast lock, the per-beat actions in order, the camera behaviour, and the short negative list from the constraints file. Then call `generate_video`:
   - `prompt`: the compressed script prompt.
   - `input_urls`: the key-art still from step 4 (first frame / reference image).
   - `model='i2v'` for the standard animate-the-still path; `model='ref2v'` only when the user wants the cast kept from supplied reference images rather than from the generat