Halloween Pet Costume — Spooky & Cute · Open Skill 公開技能

Halloween Pet Costume — Spooky & Cute

Halloween Pet Costume — Spooky & Cute 是 Digen 上的公開 Open Skill。開啟後即可了解此工作流能製作什麼。

@eggs

Auto

0 0 0

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284862/2026-10-16/tcnmwsfy.jpg

# Halloween Pet Costume — Spooky & Cute

Turns one uploaded pet photo into a photorealistic Halloween portrait: the pet wears a costume chosen for its species, breed, size and personality, standing in a fantasy Halloween scene where a roadside jack-o'-lantern suddenly lights up and the pet meets its gaze. Spooky atmosphere, cute result.

There is no structured project state. Everything needed (the pet's identity details, the chosen costume, the reference image) lives in this conversation or on the canvas — read it from the user's message, from `recognize_image`, or by reading these reference files:

- `references/style-and-prompt.md` — Style / Content / Layout / Must keep / Must avoid plus the copy-paste English prompt template. **Required before every generation.**
- `references/step-details.md` — identity-extraction checklist, costume selection matrix, pre-send QC checklist, retry rules, follow-up-edit rules.

## When to use

- User uploads a pet photo and asks for a Halloween costume, a spooky portrait, a "Halloween makeover", or names a costume (vampire / witch / pumpkin / superhero / ghost).
- User asks for "a few options" of costumes for the same pet.
- User wants to edit a Halloween pet portrait already made in this conversation (new costume, new prop, darker or cuter mood) while keeping the same pet.

Do **not** use it for plain pet photos with no Halloween theme, or when no pet is in the input (see Notes).

## Steps

1. **Read the reference photo and lock the pet's identity.**
   Use the uploaded image or asset URL as the reference. List the identity anchors you must preserve: species, breed, size/build, coat colour and pattern, eye colour, ear/tail shape, muzzle and paw markings, collar or harness, distinctive features (scar, missing ear, heterochromia, wheelchair, grey muzzle). If the photo is only available as a URL or asset and those details are not clearly visible in the conversation, call `recognize_image(image_url=<reference>, prompt="Describe this pet for identity preservation: species, breed, size, coat colour and pattern, eye colour, ear and tail shape, face markings, collar, current pose, camera angle, background and lighting.")` and work from the returned description.

2. **Choose the costume.**
   If the user named a costume, use it. If not, pick automatically the way this app does — from the pet's species, breed, size, build and apparent personality — following the selection matrix in `references/step-details.md` (call `read_skill_file(name='halloween-pet-costume', path='references/step-details.md')` now). Also read there the identity-anchor rules and the QC checklist you will use in step 5. Never dress the pet in something that conflicts with its body (no wings that hide the ears, nothing that covers the eyes or needs a human posture).

3. **Build the final prompt.**
   Call `read_skill_file(name='halloween-pet-costume', path='references/style-and-prompt.md')` and fill its prompt template: the Style lines (Halloween atmosphere — decor, props and lighting all on theme; hyper-realistic photography where skin, fur, fabric and light behave like a real photograph), the Content lines (fantasy scene: a jack-o'-lantern on the roadside suddenly lights up, the pet and the lantern meet each other's gaze, spooky-but-cute), the Layout lines, your costume description, and your identity-anchor list. Prompt language is English; a proper name supplied by the user can stay as-is. Follow the Must keep / Must avoid lists verbatim — do not add costumes, props or story beats this skill does not describe.

4. **Generate the portrait.**
   `generate_image(prompt=<filled English prompt>, input_urls=[<reference image>], model='i2i.hd.lite', orientation='square', resolution='1080P', name='<Pet name or "Halloween pet portrait">', entity_type='characters')`
   - `model='i2i.hd.lite'` is fixed: this app always redraws an uploaded pet photo. Do not switch to text-to-image and do not write a physical model name.
   - `orientation`: de