Models, splats and video
Video to 3D (experimental)
Walk through a place with a camera and get the place back as a coloured point cloud, on your own machine, with a choice of reconstruction engine. Experimental - how well it works depends on the capture.
When to use it
For a quick 3-D record of a room or a small site when no scanner is at hand. It is not a survey: there is no scale control and no accuracy statement, and the result should not be measured as if it were a laser scan.
Step by step
- Open File › Video to 3D (Experimental)….
- Choose an engine: Geometry only (the default: no model and no GPU, pure Rust, needs nothing else), Depth Anything 3 (handles shakier footage and wider turns) or VGGT (the densest result, for a room walked slowly). An engine not included in your build is shown as Not in this build.
- Press Choose a video… and pick an
mp4,mov,mkv,webm,aviorm4vfile. - Wait for the reconstruction. Long clips are processed in windows and glued with a Sim(3) pose graph, then oriented so gravity points down.
- The result opens as a point cloud: segment it, classify it, voxelise it or export it like any other.
What it was measured at
Depth Anything 3 running through the pure-Rust tract engine on the CPU takes about 23 s a frame. The default build links no C at all; the ONNX Runtime engines are opt-in builds.
Limits, stated
- Experimental. Results depend heavily on the capture: slow, steady, well-lit walks work best.
- No metric scale: the reconstruction’s size is arbitrary until you rescale it by hand.
- Which engines are present depends on the build; the default build ships the engine that needs nothing else.
Related questions
Does classification need a powerful GPU?
No. The classifiers, segmentation, decomposition and the ground filter run on the CPU, on your machine. The GPU draws the scene.
Does my data leave my machine?
No. Files are opened, processed and written locally. N3D has no upload, no cloud service and no account.
More in the FAQ.
Related guides
segment column for every point.