Gothic Halloween Cosplay Transform · عوامی Open Skill

Gothic Halloween Cosplay Transform

Gothic Halloween Cosplay Transform Digen پر ایک عوامی Open Skill ہے۔ کھولیں اور دیکھیں یہ ورک فلو کیا تخلیق کرتا ہے۔

@eggs

Other

1 1 1

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284862/2026-10-16/x22rwmwa.jpg

# Gothic Halloween Cosplay Transform

Redraws the character from the user's reference image in a gothic Halloween cosplay outfit, staged in front of a gothic cathedral graveyard gate, and can additionally produce a costume reference image and a short clip of the same character wearing that outfit. There is no structured project state: anything that must survive across turns (costume choices, stage choices, the asset being iterated on) is carried in the conversation and in the canvas assets already generated.

## When to use

- The user uploads an image and asks for a Halloween, cosplay or costume-swap version of it.
- The user asks for "gothic Halloween look", the costume reference image, or a short clip of the look.
- The user wants to change one thing (pose, prop, colour, framing) of a transformation produced earlier in the conversation.

## Inputs and outputs

- **Input**: one reference image (required — the character to redraw) plus optional text. The optional text carries costume / pose / mood / prop / framing wishes; anything not specified stays at the defaults in `references/style-and-prompt.md`.
- **Output**: primarily one transformed image. On request, also a costume reference image (the outfit as an item sheet) and a short clip of the character wearing that outfit.

## Steps

1. **Check the inputs.** If there is no reference image, ask for one before generating — do not silently invent the character. If optional text requests a costume that conflicts with the defaults, the user's request wins.

2. **Read the references before generating** (use `read_skill_file` with the exact paths):
   - `references/style-and-prompt.md` — read before the **first** generation of the conversation, and again whenever you are unsure about costume pieces, stage vocabulary, palette or prompt wording. It holds Style, Subject/content, Layout, Must keep, Must avoid and copy-paste prompt blocks.
   - `references/workflow-stages.md` — stage order, pre-flight checklist, result checklist, branching and retry rules.
   - `references/video-content-constraints.md` — read only when the user wants the short clip.
   Do not paste the long templates into your reply; they are input for the generator only.

3. **Optional identity read-out.** If the reference image is small, blurry, or the character's details are ambiguous, call `recognize_image(image_url=<reference image>, prompt="describe hairstyle, hair colour, eye colour, face shape, apparent age, current outfit, art style")` first, and name those details explicitly in the prompt so the transformation keeps them.

4. **Compose the prompt from the blocks in `references/style-and-prompt.md`**: identity block → costume block → stage block → lighting/palette block → style block, plus the negative fragment. Write the image prompt in English. Chinese/Japanese user input is only used to fill the blocks, never pasted raw into the prompt.

5. **Transform the image** (image-to-image, the reference image is the first item of `input_urls`):
   `generate_image(prompt=<assembled prompt>, input_urls=[<reference image>], model='i2i.hd.lite', orientation='portrait', resolution='1080P', name='<short title, e.g. Gothic Halloween — <character>>', entity_type='characters')`
   Always pass the channel `i2i.hd.lite`; never write a physical model name. Portrait + 1080P is the default for this poster-style look; only change `orientation`/`resolution` if the user explicitly asks (e.g. a square avatar crop → `orientation='square'`).

6. **Inspect the result** against the checklist in `references/workflow-stages.md` (identity kept, costume set complete, stage props present, no text/watermark, full body inside the frame). If it fails, retry once with a tightened prompt that names the missing or drifting elements; if it fails twice, report honestly what could not be achieved.

7. **Optional extras — only when the user asked for them**:
   - *Costume reference image* (the outfit alone, as a reference sheet): read the costume-sheet