Anime Hydraulic Press Squash · Open Skill 공개 스킬

Anime Hydraulic Press Squash

Anime Hydraulic Press Squash은 Digen의 공개 Open Skill입니다. 열어 보면 이 워크플로가 무엇을 만드는지 알 수 있습니다.

@eggs

TryOn

4 0 1

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284862/2026-10-16/97x2gspu.gif

# Anime Hydraulic Press Squash

Turn the user’s uploaded **subject** into a **2D anime illustration**, compose a **hydraulic-press keyframe**, author a **detailed video script document**, then **`ref2v`** using **the script + keyframe image** so motion and timing match the written beat sheet.

No structured project state; progress lives in conversation, canvas assets, and the script document.

**Language lock:** This file and all of `references/` are **English only**. Reply in the **user’s language**. Script beat sheets and every `generate_image` / `generate_video` **`prompt` are English**.

Do not expose channel names, tool names, or full prompts to the user.

## When to use

- User wants a **hydraulic press squash** / **flattened anime character** / **anime + hydraulic press** short video and has (or will provide) a subject image.
- User wants **anime conversion still only**, or **regenerate video** from an existing keyframe and/or script.
- User asks to change **orientation**, **pacing**, or **style** (chibi, slower press).

**Do not use:** realistic gore, humiliation targeting an identifiable real person, explicit NSFW, or exact likeness with no reference.

## Reference files (read on demand — never load all at once)

| Stage | File | Tools |
|---|---|---|
| Upload → anime I2I | `references/anime_i2i_spec.md` | `generate_image` |
| Press scene keyframe | `references/press_scene_spec.md` | `generate_image` |
| Video script document | `references/video_script_spec.md` | `create_document`, `update_document`, `get_document` |
| Reference-to-video | `references/press_video_timeline.md` | `generate_video` |
| Channels / retries | `references/generation.md` | Read before any generation |
| Pause buttons | `references/guidance-buttons.md` | Every pause turn |

## Pipeline

1. **Intake:** No upload and no **direct generate original character** → ask for one subject image (one short line).
2. **`write_todos`:** Full video request → **anime I2I → press keyframe → video script → ref2v** (four steps).
3. **`recognize_image`** (upload path): Subject type, pose, outfit, hair, expression; do not alter species or core traits in prompts.
4. **Anime I2I** (`references/anime_i2i_spec.md` + `references/generation.md`)
   - **`i2i.hd.gpt`** → **`i2i.hd`** → fallback **`i2i.hd.lite`** → **`i2i`**.
   - `entity_type`: `characters`; `name`: e.g. `Anime subject`.
5. **Press keyframe I2I** (`references/press_scene_spec.md`)
   - `input_urls[0]` = step-4 anime image; same channel ladder.
   - `entity_type`: `storyboard_images`; `name`: e.g. `Hydraulic press keyframe`.
   - **Still only / no video** → stop after step 5.
6. **Video script document** (`references/video_script_spec.md`) — **required before any `generate_video`**
   - **`create_document`** with full timed beat sheet, deformation rules, and **`ref2v prompt assembly`** English block.
   - `entity_type`: `scripts`; `name`: e.g. `Hydraulic press squash — video script`.
   - Keep `asset_id` for regen; use **`update_document`** when the user changes pacing or beats only.
7. **`generate_video` — `model='ref2v'`** (`references/press_video_timeline.md` + `generation.md`)
   - **`get_document`** if needed; **`prompt`** = script’s ref2v assembly block (English).
   - **`input_urls`**: `[step-5 keyframe]` (at least one image).
   - **`duration`** and **`orientation`** must match script **Meta**.
   - `entity_type`: `final_videos`; `name`: e.g. `Anime hydraulic press - squash`.
8. **Reply:** Brief summary; mention that the **script is on canvas** for review. Never invent URLs.

## Partial edits

| User intent | Action |
|---|---|
| Redo anime look only | Step 4; then redo 5 → 6 → 7 if video already existed |
| Redo press keyframe only | Step 5; **update_document** script if framing changed; step 7 |
| Edit script / pacing only | **`update_document`** on script; step 7 with same keyframe |
| Regenerate video only | **`get_document`** + latest keyframe → step 7 (skip 4–5 if unchanged) |
| Slower / longe