Cute Halloween Style · publiczny skill Open Skill
Cute Halloween Style
Cute Halloween Style to publiczny Open Skill w Digen. Otwórz go, by zobaczyć, co tworzy ten workflow.
Anime
0 0 1
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284865/2026-10-22/wxfaats7.jpg
# Cute Halloween Style Turn one uploaded photo into a cute, anime-inspired Halloween cosplay package: a costume reference image of the same person in a pastel Halloween outfit, followed by a short dance video of that character wearing it. Two deliverables, always tied to the same character design: 1. **Costume reference image** — the uploaded person re-dressed in a kawaii Halloween cosplay outfit. 2. **Short dance video** — the same character, same outfit, in a short playful dance clip. There is no structured project state. Keep the costume description, palette, and any user preferences as plain text in your replies so later turns can reuse them. ## When to use - User uploads a photo and asks for a Halloween / costume / cosplay / dance short video. - User asks for a cute anime version of their photo wearing a Halloween outfit. - User wants only the costume reference image (no video) — run the image stage and stop. - User wants to modify a previous result: different costume color, different Halloween prop, different dance move, longer clip, softer background. ## Inputs and outputs | | | |---|---| | Input | One reference photo of the person (required). Optional free-text wishes: costume theme (witch, cat, pumpkin, ghost, vampire), color palette, dance energy, clip length, aspect ratio. | | Output | `generate_image` asset with `entity_type='characters'` (costume reference) + `generate_video` asset with `entity_type='final_videos'` (dance clip). | If the user gave no photo, ask for one and stop. Only if the user explicitly refuses to upload a photo may you fall back to text-to-image (`model='t2i'`) — and say clearly that the result will not preserve their identity. ## Stages (references/) Details live in `references/`. Do not read them all at once — judge the current stage, then read only the file for that stage. | Stage | File | Tool / channel | When to read | |---|---|---|---| | 0. Read the photo | `references/workflow-stages.md` | `recognize_image` | Always first | | 1. Style and prompt limits | `references/style-and-prompt.md` | `generate_image` / `i2i.hd.lite` | Before writing any image or video prompt | | 2. Costume reference image | `references/workflow-stages.md` | `generate_image` / `i2i.hd.lite` | When producing or fixing the outfit image | | 3. Shot and motion limits | `references/video-content-constraints.md` | `generate_video` / `ref2v` | Before the video stage | | 4. Video script | `references/video-script.md` | `generate_video` / `ref2v` | Before every `generate_video` call | ## Steps 1. **Analyze the upload.** Call `recognize_image` on the user's photo with a prompt asking for: face shape, eye shape and color, hairstyle, hair color, skin tone, apparent age range, body proportions, current clothing, pose, camera framing, and background palette. Keep the answer as text — it is the identity anchor for every later prompt. 2. **Read the style contract.** Call `read_skill_file(path="references/style-and-prompt.md")`. Use its Style / Content / Layout / Must keep / Must avoid sections and its English prompt template to compose the image prompt. Do not invent style rules that are not in that file. 3. **Read the stage details.** Call `read_skill_file(path="references/workflow-stages.md")` and follow its Stage 2 checklist (identity anchor, costume design sheet, quality gate, single retry rule). 4. **Generate the costume reference image.** Call `generate_image` with the composed English prompt, the user's photo in `input_urls`, `model='i2i.hd.lite'`, `orientation='portrait'`, `resolution='1080P'`, `name='<short costume name>'`, `entity_type='characters'`. Run the quality gate in `workflow-stages.md`; if it fails, fix the prompt and retry once. 5. **Check the gate before video.** Confirm the costume image keeps the person's face, hairstyle, and proportions, shows the full outfit without cropping the signature props, and has no text or extra limbs. If the user only asked for the image, reply now and stop here (skip