Aesthetic Portrait · مهارة Open Skill عامة
Aesthetic Portrait
Aesthetic Portrait هي Open Skill عامة على Digen. افتحها لتعرف ماذا ينشئ هذا التدفق.
Other
2 1 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284862/2026-10-16/egffz5tg.jpg
# Aesthetic Portrait
Turns a text description of a person into one polished, "aesthetic" portrait image, and can also restyle an uploaded photo or an existing canvas asset into that same visual language while keeping the person's identity.
Scope: one request = one portrait image. There is no project state; the only durable result is the image asset on the canvas.
## When to use
- The user describes a person (identity, theme, wardrobe, pose, setting, mood) and wants an image.
- The user uploads a portrait photo and wants a more beautiful, styled version of the same person.
- The user points at a previously generated image or canvas asset and asks to re-theme, restyle or beautify it.
- The user asks for a variant of a portrait already produced in this conversation ("same person, winter theme").
Not for: non-portrait images (product, landscape, logo), multi-shot storyboards, video, or adult / NSFW content — decline politely and do not call the tool.
## Steps
1. Parse the request and extract: who the person is, the theme and mood, wardrobe, pose, setting, framing, and any feature the user explicitly says to keep (a face detail, accessory, colour, background element). Everything the user did not specify is filled from the style conventions — never invented.
2. Before any generation, call `read_skill_file(name="aesthetic-portrait", path="references/style-and-prompt.md")` and follow it. It contains the Style / Subject / Layout / Must keep / Must avoid rules, the theme preset table (wardrobe + pose + scene + light per theme), and the copy-ready English prompt templates. Do not write the prompt from memory.
3. Decide the branch and build the prompt from the matching template:
- Text only → text-to-image.
- Uploaded photo, canvas asset, or "make this image more beautiful" → image-to-image: the input image goes into `input_urls` and an explicit identity-lock clause goes into the prompt.
If the branch is unclear (several attachments, a text description plus an unrelated image, a mixed re-theme request), read `references/step-details.md` first and use its decision checklist.
4. Call `generate_image` once:
- Text only: `generate_image(prompt=<T2I prompt>, model='t2i', orientation='portrait', resolution='1080P', name=<short English title>, entity_type='portraits')`
- Image input: `generate_image(prompt=<I2I prompt with identity lock>, input_urls=[<asset_id or s3:// uri>], model='i2i.hd.lite', orientation='portrait', resolution='1080P', name=<short English title>, entity_type='portraits')`
- `name` is a short English title, e.g. `Aesthetic Portrait - Snowy Night`. Always pass `entity_type='portraits'` so the user's portraits group together on the canvas.
- Frame: default `orientation='portrait'`; use `square` for an avatar or `landscape` only when the user asks for it. Default `resolution='1080P'`; raise to `2K` only on an explicit high-resolution / print request. Never pass pixel `width` / `height`, `batch_size`, or a physical model name — `t2i` and `i2i.hd.lite` are channels.
5. Review the result with the post-generation checklist in `references/step-details.md` (identity kept on the I2I branch, one subject, no text/watermark, clean hands and face). If it is wrong, read that file again and retry with the listed fixes, at most twice.
6. Reply in one or two natural sentences: what was generated and which theme it uses. Do not paste URLs, do not dump the prompt. If the tool returned `success=false`, say so plainly and give the reason; never invent an image link.
7. When you stop at a natural point, offer the next step in the same turn with `set_guidance` — e.g. "Same person, autumn theme", "Make it a square avatar", "Warmer evening light".
## Notes
- One call produces one image; there is no batch or series mode. For several themes, run the steps again per image.
- Identity is the hard constraint of the image-to-image branch: facial features, face shape, skin tone and the person's identity must stay exactly as in t