Halloween Costume Portrait Photography · Open Skill 공개 스킬

Halloween Costume Portrait Photography

Halloween Costume Portrait Photography은 Digen의 공개 Open Skill입니다. 열어 보면 이 워크플로가 무엇을 만드는지 알 수 있습니다.

@wwws

Roleplay

0 0 1

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

# Halloween Costume Portrait Photography

Turn an uploaded photo into a studio-quality Halloween costume portrait: the same person, re-styled with themed costume, makeup, props and lighting. Reference-image editing only — nothing is generated from scratch, and the subject's identity must survive the restyle.

## When to use

- The user uploads a portrait/photo of a person and asks for a Halloween, witch, vampire, pumpkin, sugar skull, skeleton, dark fairy or "spooky glamour" version of it.
- The user wants to change costume/makeup/scene/pose of a previously generated Halloween portrait of the same photo.
- Not for: generic Halloween posters or pumpkins with no person; text-to-image scenes with no uploaded person (say what is needed instead).

## Inputs and outputs

| | |
|---|---|
| Input | At least one reference photo of the person (the identity source). The first image is the face/identity anchor. |
| Output | One still image: a Halloween costume portrait of that person. |

How the input is used: the photo is passed as `input_urls` and is the *identity* source only — facial geometry, skin tone, age, hairstyle and body type are carried over; costume, makeup, pose, props and scene are replaced according to the theme.

## Steps

1. **Check the input.** Confirm a reference photo is present and the face is clearly visible. For the intake checklist, multi-person handling and what to do with very low-quality or non-photo inputs, read `references/step-details.md`.
2. **Confirm the costume direction.** If the user already named a look (witch, vampire, pumpkin…), use it. If they only said "Halloween", offer a short menu (gothic witch / vampire countess / pumpkin / sugar skull / skeleton glam / dark fairy) and default to the gothic witch look when they don't care. Archetype details live in `references/style-and-prompt.md`.
3. **Read the visual playbook.** Before generating anything, call `read_skill_file(name='halloween-costume-portrait', path='references/style-and-prompt.md')` and follow its Style / Content / Layout / Must keep / Must avoid sections. Do not improvise the look from memory.
4. **Write the prompt.** Fill the English prompt template from `references/style-and-prompt.md` with this request's specifics: identity-keep clause, costume, makeup, pose, props, palette, lighting, framing. The prompt is fully English; it must be one concrete portrait brief, not a pile of generic keywords. Exclusions are phrased inside the prompt text (the tool has no negative-prompt field).
5. **Generate.** Call `generate_image`:
   - `prompt`: the filled English template.
   - `input_urls`: the user's uploaded reference photo (identity anchor, first item).
   - `model='i2i.hd.lite'` (image-to-image; this is the only channel used by this skill).
   - `orientation='portrait'` (head-and-shoulders vertical portrait, matching the reference framing).
   - `resolution`: omit unless the user asks for a specific quality.
   - `name`: short label, e.g. `Halloween Portrait - Witch`.
   - `entity_type='characters'`.
6. **Inspect the result.** If the face drifted, the costume was ignored, the framing is wrong, or anything in Must-avoid appeared, fix the prompt with the retry rules in `references/step-details.md` and regenerate once or twice. Never present a drifted-identity image as "the same person".
7. **Reply briefly.** One or two sentences on the look created (costume, makeup, scene), plus one practical next step. Do not paste URLs. If the tool returned `success=false`, say so plainly and state what is missing.

## Notes

- Identity is the product: a beautiful Halloween image with a different face is a failure. Always regenerate when identity drifts.
- Keep everything tasteful and non-explicit — costume is themed clothing (lace, velvet, corsets, cloaks), never lingerie or sexualized framing; the `i2i.hd.lite` channel does not support NSFW content. Gore, blood, wounds and real weapons are out as well; Halloween here means horror-glamour, not violence.
- Only