AETHER StyleSense

AI styling studio that unifies YouCam Apparel VTO & Skin AI in one session — try on garments, get your skin scored, and receive a personalized AI verdict on whether the look works/good on you.

Comment

Inspiration

Every online shopper faces the same problem: "Will this actually look good on me?" Existing try-on tools show you the garment — but they don't know your skin tone, your undertone, or your current skin condition. We wanted to build something that answered the full question, not half of it.

What it does

AETHER StyleSense is a virtual styling studio that runs two YouCam AI pipelines in a single session:

  1. Apparel Virtual Try-On — uploads your photo and a garment to YouCam's cloth-v4 generative engine, which renders the garment photorealistically onto your body.
  2. Skin AI Analysis — submits your face to YouCam's Skin Analysis API and returns scores across 9 metrics: wrinkles, pores, oiliness, redness, acne, moisture, texture, eye bags, and firmness.
  3. AI Style Verdict — synthesizes both outputs into a personalized fashion verdict: occasion fit, skin tone harmony, style alignment, and colour confidence.

The result: you see your garment, your skin data, and an AI opinion on whether they work together — in one flow.

How we built it

  • Frontend: Vanilla HTML/CSS/JS with MediaPipe Pose for in-browser body landmark detection and auto-framing
  • Backend: Node.js + Express acting as a resilient API proxy
  • YouCam APIs: cloth-v4 (task-based VTO) + Skin Analysis (async polling with wrinkle, pore, oiliness, redness, acne, moisture, texture, eye_bag, firmness actions)
  • AI Verdict: OpenAI GPT-4o-mini with rule-based fallback
  • Resilience: Three-tier fallback chain ensures the app never shows an error screen

Challenges we ran into

  • Skin AI photo requirements: YouCam Skin Analysis requires the face to occupy a sufficient portion of the frame. We solved this by auto-cropping the top 32% of any uploaded photo (the face zone for a standing person) and upscaling it to 640px minimum before submission.
  • S3 download CORS: YouCam VTO results are served from pre-signed S3 URLs which browsers block for direct download. We built a /api/proxy-image server route to fetch and stream the image with proper Content-Disposition headers.
  • Async task polling: Both YouCam APIs are task-based (create → poll → result). We built a unified polling engine with early-exit on permanent errors and graceful fallback on timeout.

Accomplishments that we're proud of

  • A working, polished end-to-end styling app
  • Successfully integrated two YouCam API tracks (Skin AI + Apparel VTO) in a single coherent UX
  • MediaPipe Pose integration for intelligent body framing before the API call
  • Zero error screens — the resilience layer always produces a meaningful result

What we learned

YouCam's APIs are genuinely powerful but require careful image preparation. The Skin Analysis API in particular taught us a lot about face detection constraints and the importance of pre-processing inputs before hitting ML endpoints.

What's next for AETHER StyleSense

  • Live camera mode with real-time garment overlay
  • Wardrobe history and saved looks
  • Skin trend tracking across sessions
  • Retailer API integration for live product catalogues

Built With

Share this project:

Updates

Submission history