AR capture on any iPhone
Corner tapping on ARKit world tracking, with a floor baseline and projection for walls taller than the camera can see. Results go into one room model shared by both capture paths (corner tapping and LiDAR).
iOS app · AR · on-device ML
Try real furniture in your room before you buy it.
Snug scans your room with any iPhone's camera, places real Amazon products in it at their true size, and tells you honestly whether each one fits, including when it's too close to call.
The demo runs in your browser with sample rooms. The iPhone app measures your real one.
3.4 cmmedian wall error vs. a tape measure
Any iPhoneAR capture, no LiDAR needed
0.74 mAP50custom furniture detector, on-device
197unit tests, run in CI
96%smaller 3D assets (682 MB to 28 MB)
How it works
Furniture returns are expensive and most come down to one question: will it fit, and will it look right? Snug answers it in three steps.
-
1
Scan your room
Tap the room's corners through the camera. ARKit turns them into walls, doors and windows. Pro iPhones can use LiDAR, but you don't need it.
-
2
Place real products
Pick from real Amazon listings. Each one drops into a 3D model of your room at its listed size, in its true colors.
-
3
Know before you buy
Every piece gets a fit verdict against your walls, doors and other furniture. Then buy the whole room in one list.
An honest fit check
A phone scan isn't a laser. Snug knows its measurements can be off by a few centimeters, so it never shows a green check it can't back up. Every verdict accounts for a ±5 cm uncertainty band.
- Fits with room to spareClear by more than twice the margin.
- FitsClear by more than the margin.
- Too close to callWithin the margin. Snug tells you which wall to measure.
- Won't fitOverlaps by more than the margin.
Engineering
A native iOS app built end to end: capture, ML, geometry, rendering and the asset pipeline.
On-device furniture detection
A YOLO model fine-tuned on merged open datasets to detect 11 furniture classes at 0.74 mAP50. It runs on the phone with CoreML, so the furniture you already own shows up in your scan.
Fit engine
Separating-axis collision on rotated footprints, plus clearance to walls and doorways, classified against the uncertainty band. It's pure, deterministic and covered by Swift Testing.
True-to-life 3D
A RealityKit diorama under neutral white light, so colors match what arrives. An offscreen renderer makes the room thumbnails and before/after share images.
3D asset pipeline
Mesh simplification and compression cut the product models from 682 MB to 28 MB, small enough to download over cellular.
Built to last
SwiftUI with MVVM, SwiftData with versioned schema migrations, no third-party dependencies in the app, zero compiler warnings, and CI on GitHub Actions.
- Swift
- SwiftUI
- RealityKit
- ARKit
- CoreML
- YOLO
- SwiftData
- Swift Testing
- GitHub Actions
- TypeScript
- three.js
Try it in your browser
The web demo runs the same fit check as the app. It's ported line for line to TypeScript, and the Swift test cases are ported with it. Drag real products around a sample room, or type in your own room's size.