Pet Halloween Costume · সর্বজনীন Open Skill

Pet Halloween Costume

Pet Halloween Costume Digen-এ একটি সর্বজনীন Open Skill। খুলে দেখুন এই ওয়ার্কফ্লো কী তৈরি করে।

@JackRhino

Roleplay

1 0 3

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284865/2026-10-21/pcg4p4kr.jpg

# Pet Halloween Costume

Turns one uploaded photo plus a short text request into a Halloween costume portrait of the exact same subject, keeps the subject's identity intact, and can additionally deliver a reference image of the costume set and a short showcase video. There is no structured project state: every deliverable is a canvas asset, and the costume decision lives in the conversation.

## When to use

- The user uploads a photo and asks to dress the subject up for Halloween (costume, cosplay, trick-or-treat outfit, "make my cat a witch", "my dog as a pumpkin").
- The user asks for a version of an already generated costume portrait (different costume, different framing, cleaner costume, matching set).
- The user asks for a picture of the costume itself, or a short clip showing the costume being worn.

Do not use this Skill for plain restyling with no costume, for multi-scene stories, or for pure video editing.

## Steps

1. **Intake.** Look at the uploaded image and the text. Note the subject (pet species/breed or person/character), coat or hair colors, markings, eye color, face shape, current pose and camera angle, and whether the photo is photographic or illustrated. If a detail is hard to read, call `recognize_image` once on the uploaded image and ask for a factual inventory (species, colors, markings, pose, framing, lighting, style). Write these as the identity anchors — they are what must survive the costume change.

2. **Plan.** If the request involves more than one deliverable (portrait + costume reference image + optional video), call `write_todos` with one entry per deliverable and keep exactly one `in_progress`.
   Then `read_skill_file` → `references/step-details.md` and follow its intake checklist and branch rules (explicit costume vs. vague costume vs. no costume given, retry conditions, quality checks, final reply and guidance wording).

3. **Costume portrait (core deliverable).** `read_skill_file` → `references/style-and-prompt.md`. Follow its Style / Subject / Layout / Must keep / Must avoid rules and fill its portrait prompt template with the identity anchors from step 1 and the costume decided in step 2. Then call:
   - with an uploaded reference image: `generate_image(prompt=<filled template>, input_urls=[<uploaded image>], model='i2i.hd.lite', orientation='portrait', resolution='1080P', name='<short costume title>', entity_type='costume_portraits')`
   - text only, no usable image: `generate_image(prompt=<filled template>, model='t2i', orientation='portrait', resolution='1080P', name='<short costume title>', entity_type='costume_portraits')`
   Never pass a physical model name in `model`; only the channels above.

4. **Costume reference image (when the user wants the costume itself).** Reuse the same `references/style-and-prompt.md`, switch to its costume-sheet template (costume only, no subject), and call `generate_image(prompt=<filled costume-sheet template>, model='t2i', orientation='portrait', resolution='720P', name='<costume name> — costume sheet', entity_type='props')`. If the user wants the sheet to match the rendered portrait exactly, instead pass the accepted portrait as `input_urls` and use `model='i2i.hd.lite'`.

5. **Showcase video (optional, only if the user asks for a clip).** `read_skill_file` → `references/video-script.md`, write a detailed English shot script for this specific subject and costume (shot list: framing, camera move, action, costume focus, light, transition), then `read_skill_file` → `references/video-content-constraints.md` and apply its limits. Then call `generate_video(prompt=<prompt built from the script>, input_urls=[<accepted costume portrait>, <costume sheet if available>], model='ref2v', orientation='portrait', duration=5, name='<costume name> — showcase', entity_type='final_videos')`. Do not call `generate_video` before the script is written, do not pass a physical model name, and do not use it for anything but a short costume showcase.

6. **Quality check and r