Halloween Makeup · Open Skill 公開技能
Halloween Makeup
Halloween Makeup 是 Digen 上的公開 Open Skill。開啟後即可了解此工作流能製作什麼。
Custom
0 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/jjf3hpa8.jpg
# Halloween Makeup Look (reference image → Halloween makeup portrait)
Turns one reference photo into the same subject wearing a Halloween makeup and costume look. Single capability, no structured project state: the source image + the generated image on the canvas are the only artifacts.
## When to use
- The user uploads a photo and asks for a "Halloween makeup / Halloween version / Halloween costume" of it.
- The user has already generated one look and wants the same photo turned into another Halloween look (vampire → witch → sugar skull …).
- The user gives only a text description with no photo (text-to-image fallback).
## Steps
1. **Get the source image.** Use the image the user uploaded in this conversation (its `asset_id` or URL). Never invent or guess a URL. If the user sent no image, go to step 5 (text-to-image fallback) or ask them to upload one.
2. **Pick the look.** If the user named a look, use it. If not, default to a classic glam Halloween look (vampire or witch), say which one you chose, and offer alternatives afterwards. Typical looks:
- glam vampire: pale matte skin, deep burgundy smoky eyes, dark red lips, subtle fangs, black velvet cape and collar;
- witch: dark green/purple smoky eyes, sharp winged liner, deep plum lips, wide-brim black hat, lace choker;
- sugar skull / calavera: white base, black eye sockets ringed with marigold petals, stitched lips, colorful floral face paint;
- cracked porcelain doll: porcelain-white skin, cracked face paint, oversized doll eyes, stitched mouth, ribbons;
- undead / zombie: grey-green pallor, sunken dark eyes, fake blood at the mouth, tattered costume.
3. **Write the prompt in English**, using this template (fill the two placeholders):
"Apply a Halloween makeup and costume look to the person in the reference image: {LOOK_ELEMENTS}. Keep everything else identical to the reference: same person, same face and identity, same pose and gesture, same subject position in the frame, same camera angle, same framing and crop, same aspect ratio, same background layout and composition. Change only the makeup, the outfit, the accessories and the lighting mood. Halloween atmosphere: {THEME_ELEMENTS}."
`{THEME_ELEMENTS}` must keep decoration, props and lighting on the Halloween theme, e.g.: carved jack-o'-lanterns, cobwebs, candles, dried leaves, fog haze, purple and orange rim light, moonlit cool tones, subtle fake blood/spider details. Keep it tasteful.
4. **Generate the image-to-image result** with `generate_image`:
- `input_urls=[<the uploaded source image>]` — always exactly this one reference image;
- `model='i2i.hd.lite'` by default; use `model='i2i.hd'` when the user asks for higher quality or finer makeup detail; use `model='i2i.hd.gpt'` when the user asks for that variant, for stronger prompt adherence, or a more creative re-styling;
- only these three `.hd*` i2i channels are used for this task. Never pass a physical model name (e.g. `qwen-image`); it is rejected;
- `name='Halloween Makeup'` (or the look name, e.g. `Halloween Vampire Look`), `entity_type='halloween_makeup'`;
- **do not pass `orientation`, `aspect_ratio` or `resolution`** unless the user explicitly asks for them: the input image's aspect ratio and framing must stay unchanged (top/bottom/left/right proportions and subject position are locked). Pass `resolution='1080P'`/`'2K'` only if the user asks for a higher-resolution version.
5. **Text-to-image fallback** (no reference photo): same prompt template without the identity/layout lock sentence, using `generate_image(prompt=..., model='t2i', name='Halloween Makeup', entity_type='halloween_makeup')`.
6. **Reply in one or two natural sentences** naming the look that was applied and that only the makeup/costume changed. The image is added to the canvas automatically — do not paste URLs. If useful, offer next steps in the same turn with `set_guidance` (e.g. "Try a sugar skull look", "Make a higher-quality version with i2i.