Mummify Yourself Split Portrait · Open Skill 公開技能

Mummify Yourself Split Portrait

Mummify Yourself Split Portrait 是 Digen 上的公開 Open Skill。開啟後即可了解此工作流能製作什麼。

@wwws

Custom

1 1 1

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284863/2026-10-20/u9kf5vaq.jpg

---

allowed-tools:
  - generate_image
  - recognize_image
---

# Mummify Yourself — Split Transformation Portrait

Turns one uploaded photo into a single portrait-oriented still image split by a hard vertical seam: the left half is the uploaded person exactly as they are, the right half is the same person wearing Halloween mummy cosplay (aged linen bandage wraps) inside a candlelit stone crypt, with a gold embossed title lockup across the top.

This Skill produces **images only**. There is no structured project state; keep decisions in the conversation and in the generated asset.

## When to use

- The user uploads a photo and asks to be "mummified", "transformed", "cosplayed as a mummy", or wants a before/after split picture.
- The user wants the same split layout but a different costume theme, a different crypt/background, a different title wording, or a different aspect ratio.
- The user wants a fix on a previous result ("the right side isn't her face", "the title is unreadable", "keep my hoodie on the left").

## Inputs and outputs

| | |
|---|---|
| Required input | exactly one reference photo (selfie or portrait) |
| Optional input | free text: costume theme, title wording/language, keep or drop the title, aspect preference, extra details to keep |
| Output | one image, `entity_type='characters'`, short canvas `name` such as `Mummify Yourself` |
| Generation channel | always `model='i2i.hd.lite'` (image-to-image with the uploaded photo as the identity reference) |

If no photo has been uploaded, ask for one and stop — this Skill never invents an identity.

## Steps

1. **Validate the input.** Confirm exactly one usable photo (not a screenshot of a chat, not a group shot with unclear subject). If several photos or ambiguous cases appear, follow the branches in `references/step-details.md`.
2. **Load the visual rules.** Call `read_skill_file` for `references/style-and-prompt.md` (style, subject, layout, must-keep/must-avoid, prompt templates) and `references/split-composition-guide.md` (how the seam, continuity and title lockup must be built). Do not write the prompt before reading these.
3. **Read the photo.** Call `recognize_image` on the uploaded image with a short prompt asking for a factual description of: face shape, eye color and shape, eyebrows, nose, lips and teeth, skin tone, hair color/length/part, visible clothing and accessories, camera angle and distance, expression, and the background. Record these as the identity anchors used later.
4. **Write the prompt.** Assemble the English `prompt` from the template and fragments in `references/style-and-prompt.md`, filled with the identity anchors from step 3 and with the user's optional text (costume tweak, extra detail, background change). On-image title text keeps the user's language; when the user gives none, use the default English gold lockup described in the reference file. The prompt must state: hard vertical 50/50 split, left half untouched original look, right half mummy cosplay, same face/pose/expression/framing across the seam, and the title position.
5. **Generate.** Call `generate_image` with `input_urls=[<the uploaded photo>]`, `prompt=<assembled prompt>`, `model='i2i.hd.lite'`, `orientation='portrait'`, `resolution='1080P'`, `name='Mummify Yourself'`, `entity_type='characters'`. Do not pass width/height and do not use a physical model name.
6. **Self-check and retry.** Inspect the result against the checklists in `references/step-details.md` (identity, seam continuity, costume plausibility, background, title legibility, no extra text). If it fails, retry at most twice, each time adjusting only the failing clause. Then follow the retry table in that file.
7. **Optional extra image.** Only when the user explicitly asks for the costume itself as a separate reference plate ("also show me the outfit"), generate a second image with `generate_image(prompt=<costume plate prompt from references/style-and-prompt.md>, input_urls=[<the finished split image>], mod