Halloween Acrylic Standee · skill pubblica Open Skill
Halloween Acrylic Standee
Halloween Acrylic Standee è una Open Skill pubblica su Digen. Aprila per vedere cosa crea questo workflow.
Doc
0 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284862/2026-10-16/th7vafw6.jpg
# Halloween Acrylic Standee Turns an uploaded photo into a Halloween acrylic standee product image: the person is re-dressed in a Halloween cosplay outfit, rendered as a die-cut clear-acrylic standee standing on a themed tabletop. On top of the standee image, the skill can deliver the costume reference sheet and a short clip of the person wearing that costume. There is no structured project state. Everything needed across turns lives in the conversation, in the canvas assets, and in the reference files below. ## When to use - The user uploads a photo and asks for a Halloween acrylic standee / acrylic stand / 亚克力立牌 / Halloween cosplay version of themselves or a character. - The user wants the costume reference sheet or the short costume clip that comes with the standee. - The user asks to redo only one part: a different costume, different props or lighting, different plate text, a different clip action. - Do not use it for generic image generation or for non-Halloween standees; those are outside this skill. ## Inputs and outputs - Input: a reference image (the photo to preserve) plus optional text (costume preference, gesture/pose, palette, short lettering, or "no clip please"). - Output: the standee image (always), the costume reference sheet (default, skipped only if the user says no), the short costume clip (default, skipped only if the user says no). - If the user sent text but no image, ask once for a photo. If they still do not provide one, generate a text-only standee with `model='t2i'` and tell them identity can no longer be preserved from a photo. ## Reference files (read on demand) | Stage | File | Tool | When to read | |---|---|---|---| | Standee artwork, style, layout, prompt fragments | `references/style-and-prompt.md` | `generate_image` | Before every image generation | | Costume clip: shots, motion, duration, bans | `references/video-content-constraints.md` | `generate_video` | Before generating the clip | | Stage order, sub-steps, checks, branches, retries | `references/workflow-stages.md` | — | At intake, and whenever a check or branch is needed | Do not read all of them at once. Read `references/workflow-stages.md` first, then the file for the stage you are on, using `read_skill_file`. ## Steps 1. Read `references/workflow-stages.md` and follow its intake checklist: confirm an image is present, parse the user text (costume, palette, lettering, clip on/off), and list the three stages with `write_todos`. 2. Read `references/style-and-prompt.md`, then build the English standee prompt from its template: the uploaded photo goes into `input_urls` and carries identity, the user's text selects the costume and any short lettering on the plate, and the file's Style / Layout / Must keep / Must avoid blocks are baked into the prompt itself. 3. Generate the standee: `generate_image(prompt=<standee prompt>, input_urls=[<uploaded photo>], model='i2i.hd.lite', orientation='portrait', name='Halloween standee - <subject>', entity_type='final_images')`. Keep `orientation='portrait'` unless the user asks for landscape; leave `resolution` unset (platform default) or pass `1080P` when the user wants a print-ready / large file. `i2i.hd.lite` is the image-to-image channel for this app; never write a physical model name. 4. Run the identity + standee check from `references/workflow-stages.md` (face, hairstyle, skin tone, age, gender, pose kept; acrylic plate edge and base readable; any lettering spelled correctly). If it fails, retry once with a strengthened "Must keep" block and a simplified scene; at most two retries, then report honestly. 5. Generate the costume reference sheet with a second `generate_image` call using the costume-sheet fragment in `references/style-and-prompt.md` — `model='i2i.hd.lite'`, `orientation='square'` (sheet layout), `name='Costume sheet - <costume>'`, `entity_type='props'`. Skip it if the user asked for the standee only. 6. Read `references/video-content-constraints.md`, then generate th