face（face 控制）

Put a Face on Your Agent

Use this skill to give an agent, persona, group, or Telegram surface a real human-looking face fast.

The job is simple:

• open a stable face listing URL
• extract a small shortlist of faces
• send the options to the user to choose
• if one is approved, reuse that link as the avatar or suggest saving it in identity

Core Workflow

1. Start from the stable URL

Default to:

https://generated.photos/faces/beautified/{gender}/young-adult

Notes:

• beautified and young-adult are the best default for this use case
• gender is the main parameter that usually matters
• the site may also accept .../young-adult/{gender}, but the canonical form resolves to .../{gender}/young-adult

2. Only trust filters the site already exposes

From the page HTML, the route families that were verified as stable are:

• head pose: front-facing, left-facing, right-facing
• race: asian-race, black-race, latino-race, white-race
• eye color: brown-eyes, grey-eyes, blue-eyes, green-eyes
• hair color: brown-hair, black-hair, blond-hair
• hair length: short, medium, long
• emotion: joy, neutral

Do not invent deeper URL combinations from memory. If the user wants narrower filtering, open the base page and follow the exact links that page emits.

3. Extract faces from HTML before using a browser

The page already contains:

• /face/... detail links
• data-image-id
• https://images.generated.photos/... preview URLs

Use a direct fetch first. Return 3 to 6 options max.

python3 - <<'PY'
import json
import re
from urllib.request import Request, urlopen

url = "https://generated.photos/faces/beautified/female/young-adult"
raw = urlopen(Request(url, headers={"User-Agent": "Mozilla/5.0"})).read().decode()

cards = re.findall(
    r'href="(/face/[^"]+)".*?data-image-id="([^"]+)".*?src="(https://images.generated.photos[^"]+)"',
    raw,
    re.S,
)
transparent = {
    image_id: thumb.replace("\\/", "/")
    for image_id, thumb in re.findall(
        r'id:"([^"]+)".*?transparent:\{thumb_url:"([^"]+)"',
        raw,
    )
}

for n, (detail, image_id, thumb) in enumerate(cards[:6], 1):
    print(json.dumps({
        "n": n,
        "detail": "https://generated.photos" + detail,
        "thumb": thumb,
        "transparent": transparent.get(image_id),
    }))
PY

4. Treat background as extraction, not routing

Do not model background as a clean route parameter.

What was verified:

• the page UI shows a background area
• the route itself does not expose a simple stable public background parameter
• transparent versions appear in embedded data as transparent.thumb_url

Rule:

• if the user wants a transparent avatar, prefer transparent.thumb_url
• if the user wants a bigger preview, open the chosen detail page and use the 512px og:image or twitter:image

5. Always keep the user in the loop

Default output should be:

• 3 to 6 numbered options
• one normal preview per option
• transparent preview too when available
• the detail page link

Do not auto-pick unless the user asks.

6. Reuse the chosen face properly

Once the user chooses one:

• propose using it as the avatar for the agent, group, Telegram profile, or similar
• if it is their long-term face, suggest storing the chosen link in identity
• keep the detail page link, not just the image preview URL

Recommended Response Shape

When you answer, keep it compact:

I found 4 options for your agent face.

1. Option A
   Detail: ...
   Preview: ...
   Transparent: ...

2. Option B
   Detail: ...
   Preview: ...
   Transparent: ...

Tell me which one you want and I’ll prepare it as the avatar.

Common Mistakes

• guessing URL variants instead of following the page's real filters
• trying to control background through the route
• showing too many faces at once
• saving only the preview URL instead of the detail link
• changing an avatar without explicit approval

External Endpoints

No other data is sent externally.