Halloween Magic Witch · skill pública de Open Skill
Halloween Magic Witch
Halloween Magic Witch es una Open Skill pública en Digen. Ábrela para ver qué crea este flujo.
Custom
0 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/484mn6qq.jpg
# Halloween Magic Witch Portrait
Turn a user-uploaded portrait photo into a **cinematic Halloween witch portrait**: identity and likeness come entirely from the reference image, while wardrobe, pose, props, setting and lighting are fully switched to a Halloween magic theme. Output is one **vertical 9:16 half-body close-up**.
Single-step task, no multi-stage flow: get the image → pick a template → assemble the English prompt → call `generate_image` → self-check → reply.
## When to use
- The user uploads a portrait photo and asks for a Halloween / witch themed cinematic portrait (costume swap, witch hat, new background, magic light effects).
- A result already exists and the user wants a different witch hat style / cloak color / magic mote color.
- The user only tweaks one item (e.g. "make the hat wide-brimmed", "change the motes to cyan-blue").
Not applicable: non-portrait images, requests for full-body or wide shots, video, or group photos. In those cases, explain honestly that this skill only produces a single-person vertical half-body portrait and do not force a run.
## Input requirements
- **One portrait photo is enough**: a frontal, evenly lit, clearly visible face in close-up or half-body framing works best.
- If the user sends several photos, pick the sharpest, most frontal one as the reference and ignore the rest.
- If there is no face in the image, or the face is too blurry to read: state the reason and ask for another photo. **Do not** force a generation.
- Likeness and identity are **entirely delegated to the reference image**: never describe facial features, age, hairstyle, skin tone or body type in the prompt, and never repeat the person's appearance back in the reply.
## Steps
1. **Get the reference image**: the user's uploaded image (`asset_id` / https / `s3://` all work). If there is none, ask for one first.
2. **Pick a template**: default is "A · Dark Gold & Leather". If the user names a palette, switch to that template; if not, produce the default first and offer the other two in the reply. The three templates **only** swap the witch hat style, cloak color and magic mote color — theme, composition, lighting, background and portrait treatment stay identical so the user can compare them side by side.
3. **Assemble the prompt**: fill in the English skeleton below with the template phrases, section by section (image prompts are always in English).
4. **Call the tool**:
`generate_image(prompt=<assembled English prompt>, input_urls=[<reference image>], model='i2i', orientation='portrait', name='Halloween Witch · <template name>', entity_type='characters')`
- `orientation='portrait'` is the vertical 9:16 framing and is **required**; never pass pixel width/height.
- Do not pass `resolution` (default 720P is used); if the user explicitly asks for more clarity, pass `resolution='1080P'`.
- Use `model='i2i.hd'` only when higher material detail is needed (this theme involves no NSFW, so there is no downside); never write a physical model name.
- Generate one image at a time; do not batch.
5. **Self-check** (see "Self-check and fixes"): if it fails, apply the matching fix and re-run once, at most one extra run.
6. **Reply**: one or two natural-language sentences covering which template was used and the key points of the image. **Do not** paste long URLs or raw JSON. If the default template was used, you may call `set_guidance` in the **same turn** with two next-step suggestions, "Switch to the 'Wine Red & Emerald' template" and "Switch to the 'Midnight Cyan' template" (do not call any other tool in that turn).
## Prompt skeleton (English)
Replace `{{HAT_AND_CLOAK}}` and `{{PARTICLES}}` according to the template and keep everything else verbatim:
```
Cinematic Halloween witch portrait, half-body close-up of the exact person in the reference image.
Identity: keep the person's identity, facial features, face shape and likeness exactly as in the reference photo; do not alter, beautify or stylize th