Inspiration Online shopping often leads to uncertainty about fit and style. We wanted to bridge the gap between e-commerce and the in-store experience by creating an AI-powered virtual try-on that not only shows users how products look on them in real-time but also provides social proof through community voting — helping people make confident purchase decisions.
What it does Real-time virtual try-on – Users can see how clothes, accessories, or other products look on their own image or live camera feed using AI.
Community voting – After trying on an item, users can submit their look for live voting and get instant, honest feedback from other community members.
Shareable results – Users can save and share their favorite looks with friends, complete with celebratory confetti effects.
How I built it Frontend: Built with Next.js / React for a fast, interactive UI. AI/ML: Used a https://yce.perfectcorp.com/api for real-time body/face mapping and product overlay. Hosting: Deployed on Vercel for seamless CI/CD and scalability.
Challenges I ran into Latency in real-time try-on: Balancing model accuracy with speed on client-side devices was tricky. Optimized by reducing model size and using WebGL acceleration.
Vote spam prevention: Had to implement rate limiting and user authentication (anonymous + optional sign-up) to keep voting fair.
Cross-browser camera access: Handling permissions and video feed formats across Safari, Chrome, and Firefox required extensive testing.
The 404 on /write: That specific route wasn’t implemented initially — learned to plan all public routes in advance.
Accomplishments that I'm proud of Smooth real-time tracking even on mid-range smartphones.
Live voting system that updates instantly and has been used by [X] beta users.
Positive user feedback on the confetti share feature — it turns a functional tool into a delightful experience.
Successfully deploying a full-stack AI product on Vercel with zero downtime during initial testing.
What I learned How to optimize on-device ML models for performance without sacrificing accuracy.
Designing a real-time UI with optimistic updates for votes and image rendering.
The importance of graceful fallbacks (e.g., static upload option if camera fails).
Serverless function limitations with long-running ML tasks — ended up using a separate media processing queue.
What's next for second-option Multi-product try-on (e.g., outfit + accessories simultaneously).
Personalized recommendations based on past votes and try-ons.
Merchant integration – Allow brands to upload their catalog and embed our try-on widget.
Mobile app using React Native for better camera performance.
Moderation system for image and vote quality.
Fix the /write route to publicly document the project’s journey! 😄
Log in or sign up for Devpost to join the conversation.