-
-
Windows 95 Theme
-
Y2K Theme (Optional)
-
-
-
-
-
Inspiration
Y2K fashion is back, and the best pieces are second-hand on Depop, Vinted and eBay, where you can't try anything on. We wanted you to see the look on yourself first, then find the closest possible item second-hand.
What it does
For Our first hackathon we made RetroFit, an iPhone app for trying on Y2K fashion. The camera tracks your face and puts 3D glasses, hats and accessories on you live. For clothes, an AI model renders you wearing the outfit from a photo. One tap then searches Depop, Vinted or eBay for the piece. The app has two themes: Classic Windows 2000 or Y2K Chrome.
How we built it
- The app is native SwiftUI, with its own Win95-style design system using AI and Basic UI design to make.
- Live try-on uses ARKit face tracking with RealityKit.
- We made the hats, the accessories and most of the glasses ourselves in Blender using the Blender Mcp with Claude Code.
- AI 'try-on' runs on fal.ai, behind our own Vercel function in TypeScript, ensuring the AI key never lives on the phone.
Challenges we ran into
- Blender's export made see-through plastic solid - fixed it by writing the transparency back into the 3D files.
- Every model has to fit any face, so we built them all around one reference head and scale each to your eye distance.
- Using the current Fal.ai models available to us, sometimes loading generated outfit designs could take 1-3minutes
Accomplishments that we're proud of
- A working AR live try-on with items publicly available or made with Blender's Mcp
- Two complete themes.
- A working connection between the app and Fal.ai using Vercel to generate higher quality outfits styles when AR wasn't suitable.
What we learned
How ARKit and RealityKit fit 3D models to a face, how to keep an API key safe behind a serverless function, and overall swift development in Xcode
What's next for RetroFit
Full-body AR try-on for clothes, (expecting huge challenges in cloth simulation physics).
Log in or sign up for Devpost to join the conversation.