Wes Bos Turns iPhone LiDAR Scans Into Editable Three.js Scenes

Building a browser-ready 3D room from a phone scan commonly requires separate capture, mesh-cleanup, and engine tools. Wes Bos has released 3D Model Anything, a project that combines an iPhone capture app with a coding-agent skill that generates Three.js source. The repository has attracted 1.1k GitHub stars.

The repository includes the capture app, a Three.js viewer, the scan-to-threejs skill, helper scripts, and a reconstructed office. Developers can inspect the result in the live office demo before cloning the code.

A scan becomes source code

The capture app records LiDAR depth measurements, camera data, and reference photographs. After export, a compatible coding agent follows the local skill’s instructions to inspect those inputs, model the room, render matching camera views, and revise the geometry.

1. Scan an interior with a LiDAR-equipped iPhone or iPad.
2. Export and extract the capture folder.
3. Open the repository in an agent that supports local skills.
4. Ask the agent to reconstruct the room with scan-to-threejs.
5. Review the generated scene and repeat the visual comparison process as needed.

Matched-camera renders help constrain dimensions, object placement, and perspective. The agent compares each render with the corresponding photograph, then adjusts the Three.js scene to improve alignment. Reconstruction therefore takes multiple passes and consumes both inference tokens and elapsed time.