IMG2.5 Candid Street-Style Portrait · सार्वजनिक Open Skill

IMG2.5 Candid Street-Style Portrait

IMG2.5 Candid Street-Style Portrait Digen पर एक सार्वजनिक Open Skill है। खोलें और जानें कि यह वर्कफ़्लो क्या बनाता है।

@eggs

Manhua

0 1 0

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284862/2026-10-16/p4rfqca2.jpg

# IMG2.5 Candid Street-Style Portrait

Generate one candid street-fashion photograph of a single stylish adult woman from a text description plus an optional reference portrait. The reference controls identity; clothing, accessories, pose and street scene can change with the text. Output is always a single image in a vertical 3:4 frame.

## When to use

- The user uploads a portrait and wants a candid street-style / everyday street-look photo of that same person.
- The user describes a woman, outfit and street moment and wants this style of image without a reference.
- The user asks to change only one part of an already generated street portrait (swap the outfit, headphones, bag, prop or street location) while keeping the identity and the overall style.

## Inputs and outputs

| Input | How it is used |
|---|---|
| Text | The only source for outfit pieces, headphones, bag details, prop and the street moment. If the user's text is thin, pick one coordinated everyday outfit + street moment yourself and keep it coherent; do not silently change the subject. |
| Reference image (optional) | Identity anchor. Face shape, features, hair and age come from it and must be preserved. Clothing, pose and scene are replaced per the text / per the style file. |

Output: one image (produced by `generate_image`, automatically added to the canvas). No text overlays, captions or collage panels are part of the output.

## Steps

1. Extract the styling brief from the text: outfit pieces, headphones, bag + charm details, prop, street moment, pose. Keep it a coordinated everyday look, not a mix of unrelated items.
2. If a reference image is attached, treat it as the identity source. If it is not directly visible in context, call `recognize_image(image_url=<the reference asset_id / uri / https url>)` with a prompt like "describe the face, hairstyle, age and build of this person for identity preservation", and reuse that description as the identity clause.
3. **Before generating, always** call `read_skill_file(name="candid-street-style-portrait", path="references/style-and-prompt.md")` and follow its Style / Subject / Layout / Must keep / Must avoid rules and its copyable prompt template.
4. Assemble the English prompt from that template, filling the outfit, headphones, bag, prop and street moment from step 1 (or the defaults in the style file when the user is vague).
5. Call `generate_image` with the channel matching the input:
   - Text only: `generate_image(prompt=<assembled prompt>, model='t2i', orientation='portrait', aspect_ratio='3:4', name='Street look portrait', entity_type='characters')`
   - Text + reference image: `generate_image(prompt=<assembled prompt>, input_urls=[<reference asset_id / uri / https url>], model='i2i.hd.lite', orientation='portrait', aspect_ratio='3:4', name='Street look portrait', entity_type='characters')`
   - `prompt` is always English. Do not pass `resolution` unless the user explicitly asks for a specific quality; one image per call.
6. If `success=false`, tell the user the generation failed and what the error said. Do not retry with a different subject and do not invent image links.
7. Reply in one or two natural-language sentences covering the outfit, accessories and street moment that were rendered. Do not paste long URLs.

## Notes

- Preserve identity: never change the face, features, hair or apparent age of the reference person. Only outfit, accessories, pose and scene are re-styled.
- If the user gives no text but a reference image, use the style file's default outfit, headphones, bag, prop and street moment instead of asking a long list of questions.
- If the user wants a single change to a previous result, re-read the style file and regenerate; do not silently redesign everything else.
- Refuse NSFW / nudity / sexualised or degrading requests politely without calling `generate_image`.
- The frame is fixed by the application: vertical 3:4 (`orientation='portrait'` plus `aspect_ratio='3:4'`).
- Only `generate