Image: GitHub
Why it mattersA room captured on a phone becomes web geometry. An agent reads the measurements and photos. It writes the scene code for you.
A room captured on a phone can now be a web page. A coding agent writes most of the code. Wes Bos opened the repository for 3d-model-anything on 21 October 2026. The GitHub API reports 667 stars on 23 October. Trendshift ranks the project 14 on its daily list the same day.
What the project is
The README calls it a workflow, not a one-click converter. You scan a room with a LiDAR iPhone or iPad. You export the capture through the share sheet. You drop the files into a local data/captures-in/<room>/ folder. From there, Wes Bos writes, you "ask an agent that supports local skills" to use the scan-to-threejs skill. The skill guides the agent through "measurement, modeling and matched-camera visual comparisons". The output is an editable Three.js scene that opens in a browser.
The viewer offers five modes for the finished scene. They are dollhouse, walking, overhead, wireframe and knolling. There is also a first-person playground. The source code and the skill are MIT. Office textures use separate asset terms. The only room in the repo is the author's office. A user starts from one worked example and brings their own scans.
What it needs on the machine
Node.js 22.12 or newer runs the viewer. Python 3 handles helpers. Pillow is used for photo comparison. Open3D with NumPy are optional for depth reconstruction. A LiDAR iPhone or iPad is needed only for new captures. The repo notes a non-LiDAR phone can still import and review scans somebody else took.
Why the number is a signal
The project is two days old on the day of this post. 667 stars in two days is fast, with a Trendshift daily rank of 14. Other developers have picked it up quickly. Wes Bos is a well-known JavaScript teacher. Some of the attention is a personal following. The workflow still has to pass a test on a reader's own room. The README says the result is "iterative". A user will edit, re-run, and compare against the photos a few times before the scene matches.
Three things the README is careful to say the project is not. It is not a scan-to-perfect-model converter. It is not a cloud service. The iPhone app is source-available. It is not on the App Store yet. A non-developer cannot install it from a public store today. A team that wants to try the pipeline has to build the iPhone app from source or wait.
The practical bar for this is simple. Can an agent rebuild one room from a scan and photos faster than drawing it by hand? If yes, the skill saves time on every room after the first. If no, a user keeps the capture and uses a modeller as before. The repo is the test, not a claim.
Source
GitHub: wesbos/3d-model-anything. Trending rank from Trendshift.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.