Pet Halloween Costume · Open Skill 公开技能
Pet Halloween Costume
Pet Halloween Costume 是 Digen 上的公开 Open Skill。打开后即可了解该工作流能制作什么。
Other
2 0 2
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284863/2026-10-19/ajthrxut.jpg
# Pet Halloween Costume Turn one uploaded pet photo (plus optional text) into a single Halloween costume portrait: the same animal, in a complete Halloween costume set, rendered as a vivid anime-style illustration on a Halloween night backdrop. Output is a still image only. ## When to use - The user uploads a pet/animal photo and wants it in Halloween costume. - The user asks for a "Halloween version", "trick-or-treat outfit", "costume set" of their pet. - The user wants an existing pet Halloween image redone with another costume, color, pose or background. ## Steps 1. **Lock the pet's identity first.** Look at the uploaded photo and write down: species, breed look, fur colors and pattern, markings, eye color, ear shape, muzzle/face shape, body proportions, current pose and camera angle. If the photo is small, dark, or ambiguous, call `recognize_image` with a short prompt asking for exactly these attributes. Follow the inspection checklist and the per-species branch table in `references/step-details.md` (read it now, it is short). 2. **Read the visual rules.** Call `read_skill_file` for `references/style-and-prompt.md` and keep its Style / Subject / Layout / Must keep / Must avoid blocks plus the English prompt template as the base for every prompt you write. Do not invent a different look. 3. **Choose one complete costume set.** If the user named a costume (pumpkin, witch, vampire, ghost, bat, skeleton, spider, candy, pirate, mummy…), use it. If the user gave no preference, read `references/costume-catalog.md` and pick one set that suits the animal's size, coat and silhouette; state the choice to the user in one short line. A "costume set" always means headpiece + body piece + collar/harness + at least one accessory, all on the same motif. 4. **Write the prompt** in English by filling the template from `references/style-and-prompt.md`: pet identity slot, costume slot, pose slot, Halloween night background slot, lighting/style slot, and the must-keep / must-avoid lines. Keep the costume anatomically fitted to a four-legged animal, not to a human body. 5. **Generate the image.** Call `generate_image` with: - `prompt`: the assembled English prompt from step 4, - `input_urls`: the user's uploaded photo (the pet photo asset; put it first if several images), - `model='i2i.hd.lite'`, - `orientation='portrait'`, `resolution='2K'`, - `name`: short title such as "Mochi the Pumpkin Cat", - `entity_type='characters'`. Only if the user explicitly asks for a costume concept with no pet photo at all, use `model='t2i'` and keep everything else the same. 6. **Quality check before replying.** Run the checklist in `references/step-details.md` (identity preserved, costume complete and fitted, Halloween theme visible, no text or watermark, no extra limbs). If it fails, fix the failing prompt line and regenerate once; never present a failed result as success. 7. **Reply** in two or three natural sentences: which costume set was used, that the pet's markings and eye color were kept, and offer one concrete next variation (another costume, lantern-lit street, close-up portrait). Do not paste URLs. ## Notes - Never change species, breed, fur pattern or eye color; the costume goes on top of the animal, it does not replace it. - Do not humanize the pet into a person; keep paws, muzzle and anatomy intact. - Do not invent details about the photo that you cannot see; if the costume request is unclear, ask one short question or pick the catalog default. - This skill produces a still image only; do not promise or call video generation. - No gore, wounds, sexualized or otherwise unsafe content: politely refuse and do not call the generator. - If `generate_image` returns `success=false`, say so plainly and stop; never fabricate a result or a link.