Skin-Aware Stylist

What it does

Most virtual try-on tools answer "does this garment fit?". Skin-Aware Stylist answers "which garment is good for me today?" You give one close portrait. YouCam AI Skin Analysis scores four skin signals (radiance, redness, dark circles, moisture). A transparent rule engine turns those scores into a ranking of a small garment catalog (color tone, brightness, texture against each result), and YouCam AI Clothes Virtual Try-On renders the top three looks on the same photo. Each look carries a one-sentence reason, for example "your redness is high, so this calm cool green avoids saturated red near the face".

Consumer and eCommerce value

• Shoppers: fewer wrong-color returns; a styling reason instead of a bare render.
• Retailers: the catalog ranking is a drop-in for a product page ("best for your skin today" sort), and the try-on image is the conversion nudge.
• One photo, two YouCam APIs, about 15 units per full session (9 skin + 3 x 2 try-on).

How it uses the YouCam API

1. File API (upload) -> AI Skin Analysis (SD, 4 concerns, 9 units).
2. File API -> AI Clothes Virtual Try-On (cloth-v4, 2 units per look) for the three garments ranked first. The API key stays on the server. A persistent unit ledger refuses any call that would exceed a safety cap, so a public demo cannot burn the free units. A key-free demo mode replays recorded real YouCam responses for the bundled sample faces, so judges can run the project without any key.

Privacy

The photo is sent only to the YouCam API (Perfect Corp keeps data up to 30 days per its terms), is held in memory only, and is deleted when the user presses "Delete my session" or after 10 minutes. Consent is required before upload. All demo faces are synthetic.

Built with / AI disclosure

Node.js (no dependencies), vanilla JS, YouCam API. Code and tests were written by Claude (Anthropic). The entrant reviews the finished work and submits it. The three demo faces are public-domain StyleGAN images from Wikimedia Commons, extended downward with the YouCam image-extender API, and the seven garment photos were generated with the YouCam image-generator API, so no real person appears. The demo video was recorded from the running app, and the narration is a synthetic voice (Google Chirp 3 HD). The stylist is a rule engine, not an LLM.

Try it

Repository (private, shared with [email protected]): see the Devpost repository field. npm start runs demo mode with no key; node --test runs 65 tests against a local mock; with a YouCam key the same page calls the live API. Full instructions are in README.md.

Honest limits

The rule engine is a transparent heuristic over four skin signals and a seven-garment catalog, not a clinical or fashion model. Try-on results come from the YouCam model and slightly re-render facial skin texture. Skin analysis needs a close portrait (face about 60 to 80 percent of the width). The project is not deployed to a public URL; judges can run it locally in demo mode.

What's next

HD skin analysis for more concerns, a retailer JSON feed, tone analysis for the palette.

Built With

• javascript
• node.js
• youcam-api