Nido seeks to make 3D room design more accessible to people of all kinds, not just technical experts. Scan your room with an iOS app, turn it into a gaussian splat, and design your space on the web with real dimensions. Created as my thesis at ITP NYU, advised by Luisa Pereira. Try it at nido.place.
Nido pairs a mobile scanning app with a room editing website. The iOS app uses Apple RoomPlan, Object Capture, and ARKit to capture rooms and objects, which are processed into 3D Gaussian Splats. On the web, spaces are edited in a Three.js / React Three Fiber editor with real-world dimensions. Uploaded floorplans also become editable 3D structures through AI agents, and product images can be transformed into 3D objects.
PROCESS
Rooms are captured on iPhone with a guided sweep using LiDAR or video, then uploaded to the cloud, where formats are converted, splats are trained, and furniture is generated. The result is served back to an editable, to-scale 3D workspace with floorplans in the browser.
TRAINING
After the thesis I rebuilt the training pipeline around measurement. Every scan now holds back a set of frames, and each run is scored on how well it predicts them and how closely its depth matches the phone's LiDAR. That made it safe to halve the training schedule: the wait for a finished room dropped from about 47 minutes to about 23 with no visible loss, and switching the delivered file to SPZ cut it from 60 MB to about 6 MB. One method that scored higher on paper but looked worse side by side was turned down.
PRESENTATION
Watch my full thesis presentation below, and read more in the ITP Thesis Archive 2026.
Live 3D scan
Real-time Gaussian splat
STILL
A bedroom scanned with the Nido app on an iPhone and trained on the 23-minute schedule. The file is 3 MB.
This is the real Nido editor, running inside this page. Launch it and choose Sample Room to explore a demo. It is built for a computer, and on a phone it says so first.