Outfit Swap · skill pubblica Open Skill
Outfit Swap
Outfit Swap è una Open Skill pubblica su Digen. Aprila per vedere cosa crea questo workflow.
TryOn
0 2 1
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284863/2026-10-20/v4qt9ukf.jpg
# Outfit Swap Take a reference photo of a person and make a new image in which the same person wears a different garment. The deliverable is a single image; there is no structured project state to maintain. ## When to use - The user uploads a photo of a person plus a photo of a garment worn by someone else and asks to swap, dress, change clothes, or "try on". - The user uploads one photo of a person and describes the target garment in text. - The user asks to redo, adjust, or fix an earlier outfit-swap result. - The user asks for a virtual try-on / look-change image built from an uploaded picture. ## Steps 1. Confirm the inputs before generating: identify which uploaded image is the person (the identity and composition source) and which is the garment. If no photo of a person exists, ask for one — never invent a person. 2. Call `read_skill_file` for `references/workflow-stages.md`. Follow the input-triage branch, the per-image inspection checklist, and the retry rules written there. 3. Inspect the references with `recognize_image` (`image_url` plus a short `prompt`). For the person photo ask about pose, framing, crop, background and light direction; for the garment photo ask about garment type, colour, pattern, collar, sleeves, length and fastenings. Use the answers only to write a more precise prompt; never show raw tool output to the user. 4. Call `read_skill_file` for `references/style-and-prompt.md` and `references/prompt-templates.md`. Compose the English `prompt` from those conventions plus the current user intent — no generic filler text. 5. Call `generate_image` for the image-to-image swap: - `model='i2i.hd.lite'` - `input_urls=[person_image, garment_image]` — person first, garment second; pass only the person image when the garment is given in text - `prompt` = the composed English swap prompt (garment, fit behaviour, must-keep, must-avoid) - `orientation='portrait'` by default; switch to `landscape` when the person photo is clearly horizontal or the user asks for a horizontal frame - `name` = a short canvas title (for example "Linen shirt swap") - `entity_type='outfit_swaps'` - leave `resolution` unset unless the user asks for extra detail (the default is fine) 6. Review the result against the must-keep / must-avoid list in `references/style-and-prompt.md`. If identity, garment structure or original framing broke, regenerate once with the fix stated explicitly in the prompt, then report honestly if it still fails. 7. Reply in one or two natural sentences about what changed. Do not paste URLs or long parameter dumps. In the same turn, offer relevant next steps with `set_guidance`. ## Notes - Keep the face, hair, skin tone, body proportions, pose, hands, accessories, background, camera angle, crop and aspect ratio of the person photo intact; only the clothing changes. - Preserve garment specifics: colour, print, logo, collar shape, sleeve length, hem, buttons, zips and trim. - Do not add garments, layers, props, text or watermarks that the user did not ask for. - Adult or explicit requests must be refused politely without calling `generate_image`; keeps to ordinary, fully clothed fashion try-on. - If a tool returns `success=false`, say so and stop; never fabricate a result or a link. - Generate one image per request; re-run only when the user asks or when step 6 requires a single correction pass. ## Reference files | File | Read it when | |---|---| | `references/workflow-stages.md` | Before inspecting images — input triage, checklists, retry and completion rules | | `references/style-and-prompt.md` | Before every generation — style, subject, layout, must-keep and must-avoid rules | | `references/prompt-templates.md` | Before every generation — copy-ready English prompt blocks to fill in | Read these files one at a time with `read_skill_file`; do not assume their contents.