AI virtual try-on, from model benchmarks to the App Store
- Client
- Stylr
- Period
- Apr 2026 – present
- Role
- Founder and sole engineer; design with a product designer

Problem
Shoppers can't see how clothes look on them; stores lose conversions and eat returns. Existing try-on tools are slow, expensive or ugly.
What I built
- Try-on lab that benchmarks generation models side by side (Wan, Qwen Image, FLUX, SeedVR2 upscaling, SAM 3 garment cleanup, Gemini) and picks the pipeline per garment type.
- Fastify API owning auth, data and jobs (Clerk, Drizzle, Neon): closet, outfits, feed, credits, daily look.
- Expo mobile app (iOS, on the App Store), Next.js marketing site, embeddable widget for partner stores, admin analytics and job inspector.
- Studio mode that prints SVG designs from my own catalogue (SVG Otter) onto garments.
Result
Live on the App Store (version 1.3, English and Croatian). Beta testing in Croatia; widget ready for store integrations.
Screens




Stack
- Expo / React Native
- Next.js 15
- Fastify 5
- TypeScript
- Drizzle
- Neon Postgres
- Clerk
- S3
- DashScope / fal.ai
- Railway
- Vercel