Soufflé | Devpost

GIF
Landing page
GIF
Marketplace
Adding a New Product
New Product on Marketplace
Cart
Multiple furniture AR
Augmented snapshot with decor suggestions

Inspiration

View items in Augmented Reality (AR), create 3D models from photos, and preview multiple pieces in your home instantly.

Online furniture shopping never feels accurate. Photos don’t show scale, dimensions are hard to picture, and you never really know how something will look in your home. We also noticed there’s no place where you can view multiple items at the same time in AR. Every tool only lets you place at most one object. We wanted a way to actually design a room, not just preview a single chair.

What it does

Soufflé is a 3D-first furniture marketplace that lets you visualize anything in your space.
Users can:

browse furniture listings with auto-generated 3D models
create listings by uploading a few photos
upload product photos from any website to preview a 3D model of it in AR
scan a generated QR code to conveniently view AR in their mobile browser
place multiple items in AR at once, something you can’t do anywhere else
view the product's dimensions live in AR
upload a screenshot of their AR space for a lighting-accurate rendering and interior design feedback

How we built it

We built the interface in React and used Three.js with WebXR for all 3D and AR rendering.
User images are sent to Meshy’s image-to-3D API, which returns a 3D GLB model. We wrote our own logic to scale models based on real-world dimensions and built a system that supports placing several objects in the same AR session. Everything runs through a lightweight flask backend that manages uploads, model generation and assets.

Challenges we ran into

Making multiple models appear in one AR scene without breaking tracking
Scaling objects correctly so they match the real size
Integrating Three.js into a React workflow without conflicts
Designing a UI that stays simple even with a complex pipeline behind it
Properly mapping colours to the generated 3D model

Accomplishments that we're proud of

A working pipeline from images → 3D model → AR
Multi-item AR support, which we couldn’t find in any existing tool
A clean React interface that still handles real-time 3D
Turning random web images into usable AR models

What we learned

We learned how WebXR sessions behave on different devices, how to handle Three.js scenes efficiently, and how to properly scale 3D models based on user dimensions. We also learned how many small details go into making AR feel natural.