Look Roulette

Your face is a slot machine. Spin it. 🎰

Comment

Inspiration

Beauty shopping has a decision problem: endless shade swatches, no clear answer to what suits your face. We built the opposite of a clinical shade-matching utility — a game you cannot put down, where your face becomes the slot machine and every spin renders a real makeup look on your selfie.

What it does

  1. Upload a selfie → YouCam Skin AI analyzes skin tone, undertone, and skin values (presented as a "Skin Analysis Report"; informational only, no medical claims).
  2. A palette generator selects hue families harmonized to your undertone.
  3. Press SPIN → YouCam Makeup VTO renders the look on your face (arbitrary hex shades with matte/satin/shimmer/metallic/gloss textures). Occasional 1-in-8 jackpot spins use Look VTO for a complete curated look.
  4. Swipe like/dislike → an SQLite taste profile makes each spin visibly smarter; Style IQ climbs on screen — the self-learning styling assistant requested by the brief.
  5. Shop This Look → mock cart checkout.

How we built it

  • YouCam APIs: Skin Analysis (/s2s/v2.0/task/skin-analysis), Makeup VTO (/s2s/v2.0/task/makeup-vto), Look VTO (/s2s/v2.0/task/look-vto) — asynchronous task endpoints with status polling. Skin AI outputs (ui_score/raw_score, heatmaps, skin_age) drive palette generation; the APIs compose rather than sit side by side.
  • Backend: FastAPI with /spin, /swipe, /looks, /cart, and /health.
  • Frontend: Streamlit casino-glam interface with a three-second spin-and-reveal hook.
  • Learning loop: deterministic undertone-to-hue-family generator plus SQLite taste tables; every swipe re-weights future palettes.
  • Demo: 101-second cinematic video constructed from genuine API renders.

Challenges we ran into

  • YouCam calls consume real API units (Skin Analysis = 9, Makeup VTO = 1), so we built a deterministic simulated mode for practice runs and reserved live calls for the demo.
  • Skin Analysis requires the face to occupy more than 60% of image width; we corrected the demo portrait framing to satisfy this constraint.
  • The test browser blocked localhost, so demo frames were rendered using the application's own UI code.

Accomplishments that we're proud of

  • A judge can feel the product in three seconds: selfie → spin → transformed face, rendered by real APIs.
  • The learning loop is visible on screen as Style IQ increases — a live demonstration of the requested self-learning styling assistant.
  • The 101-second demo shows the complete loop with genuine YouCam renders; no mockups.

What we learned

  • Measured skin data outperforms guessed shades: feeding Skin AI output into palette generation makes virtual try-on feel personal rather than generic.
  • In hackathons, a three-second emotional hook outperforms a three-minute feature list.

What's next for Look Roulette

  • Deploy a public Streamlit link so judges can spin their own selfies.
  • Expand palettes with hair and hair-color VTO; add verified-look sharing for friends.
  • Integrate a real e-commerce cart from shop-the-look to checkout.

Built With

  • fastapi
  • python
  • sqlite
  • streamlit
  • youcam-look-vto
  • youcam-makeup-vto
  • youcam-skin-ai
Share this project:

Updates

Submission history