Photogrammetry
From photos to a measured 3D model, free and on your computer
Photogrammetry turns photos into 3D, but setting it up is hard. With reverse3d you ask your agent in one sentence: it analyzes your photos, reconstructs with COLMAP, builds the mesh and gives you the dimensions and a render.
The problem
- Photogrammetry tools have dozens of parameters and a steep learning curve.
- Many alternatives upload your photos to the cloud or charge per model.
- You find out the photos are not good enough after hours of processing.
What you get
- Up-front analysis of sharpness, exposure and overlap, with which photos to retake.
- SfM + dense MVS reconstruction on GPU (CUDA) or CPU.
- A clean, measured Poisson mesh, ready to print (STL).
- Automatic render and a report.json with all the data.
How to do it with your AI agent
- 01
Take the photos
Walk around the object with good light and overlap between consecutive photos.
“Analyze the photos in mug/ and tell me which ones to retake.”
- 02
Ask your agent
One recipe runs the whole flow in a single call.
“Turn the photos in mug/ into a 3D model and tell me its dimensions.”
- 03
Use the result
model.stl, render.png, point cloud and report.json in the output folder.
“Render mug/model.stl in metallic blue.”
From the terminal
reverse3d run photos_to_model mug/ output/Start in 1 minute
Paste this sentence into Claude Code, Cursor, Copilot, Gemini CLI, Codex or Windsurf:
Install reverse3d from https://github.com/rellytechgame/Reverse3D-tool following its AGENTS.md. You have my permission to run the installer; don't ask me any questions.
Frequently asked questions
How many photos do I need?
20 to 60 is usually enough for small objects, with 60–80 % overlap between consecutive photos. reverse3d warns you if views are missing.
Do I need a GPU?
No. An NVIDIA GPU (CUDA) makes it faster; without one it runs on the CPU without failing.
Are my photos uploaded anywhere?
No. Everything is processed locally.