Closet Wizard — Your AI-Powered Smart Wardrobe

Scan your closet, build AI outfits, and virtually try them on your own photo — powered by Perfect Corp API. Style smarter. Shop only what you need.

Comment

InspirationMost people wear less than 20% of their wardrobe — not from lack of clothes, but lack of clarity. We wanted to build a tool that makes every item in your closet work harder, using AI to surface combinations you never thought of and show you exactly how they look before you get dressed.What it doesClosetAI is an AI-powered wardrobe assistant. Upload photos of your clothes and the app automatically catalogues each item — category, color, fabric, and style tags — with no manual entry. An LLM then generates complete outfit combinations with style reasoning. Users can virtually try on any outfit on their own photo using Perfect Corp's AI Clothes, Shoes, Bag, and Hat APIs. A style gap analysis identifies missing wardrobe pieces and recommends products to fill them. Every result is shareable as a styled card or animated video reel.How we built it•Frontend: React 19 + Tailwind 4, Memphis-inspired design system•Backend: Node.js + tRPC + Drizzle ORM on TiDB•AI Vision: LLM multimodal analysis for automatic clothing categorization on upload•Outfit Engine: LLM-powered combination generator with occasion, season, and style reasoning•Virtual Try-On: Perfect Corp AI Clothes, Shoes, Bag, and Hat APIs via server-side polling•Storage: S3-backed file storage with presigned public URLs for Perfect Corp API compatibility•Auth: Manus OAuth with per-user wardrobe and try-on history persistenceChallenges we ran intoThe biggest technical challenge was integrating Perfect Corp's virtual try-on APIs correctly. Their external servers require publicly accessible image URLs — our internal storage proxy paths had to be resolved to presigned S3 URLs before each API call. We also discovered that the Shoes, Bag, and Hat APIs require a mandatory gender parameter not prominently documented, which caused silent failures. Polling timeouts and transient TLS connection drops required retry logic with exponential backoff.Accomplishments that we're proud ofWe built a full end-to-end pipeline — from raw photo upload to AI-catalogued wardrobe to virtually tried-on outfit — in a single cohesive experience. The zero-manual-entry wardrobe scanning using LLM vision is seamless, and the combination of four distinct Perfect Corp APIs (clothes, shoes, bags, hats) layered sequentially onto a single user photo produces genuinely impressive results.What we learnedIntegrating third-party AI APIs in production requires careful attention to URL accessibility, required vs. optional parameters, and realistic timeout budgets. We also learned that the best AI features are the ones users never have to think about — automatic categorization on upload made the entire experience feel effortless.What's next for ClosetAI — Your AI-Powered Smart Wardrobe•Outfit of the Day — a daily AI-generated look delivered as a push notification•Multi-user style profiles — share your closet with a stylist or friend for collaborative outfit building•Retailer integrations — direct add-to-cart from style gap recommendations•Mobile app — native iOS/Android for in-store scanning with the camera

Built With

Share this project:

Updates