Guide

How to run TripoSR on a Mac

The honest DIY path, the Apple Silicon caveats, and the one-click alternative.

You can run TripoSR on a Mac. TripoSR is a small, MIT-licensed image-to-3D model from Stability AI and Tripo, so it runs on Apple Silicon, but the official code is written for NVIDIA GPUs and needs some coaxing on a Mac. Below is the manual route, the gotchas that trip people up, and the one-click way to run the exact same model fully offline inside 3D Mate.

The manual route

1

Clone the repo and make a Python environment

Clone the TripoSR repository from GitHub and create a clean Python 3.10 or 3.11 virtual environment, so its PyTorch build does not collide with anything else on your system.

2

Install PyTorch and the requirements

Install a recent PyTorch (the Apple Silicon build ships the MPS backend), then the project's requirements. This pulls in transformers, trimesh, and the rest.

3

Deal with the mesh extractor

TripoSR's default marching-cubes step (torchmcubes) is built for CUDA. On a Mac you swap it for a CPU implementation such as PyMCubes, or run extraction on the CPU. This is the step most Mac setups get stuck on.

4

Run it on a photo

With a cut-out image, run the inference script (for example python run.py your-image.png --output-dir output/). Expect a draft mesh in roughly under a minute on Apple Silicon, longer on the CPU path. A clean background cutout matters a lot for quality.

There is no CUDA on Apple Silicon, so the parts of TripoSR that assume an NVIDIA GPU either fall back to CPU or need a substitute. None of it is impossible, it is just an afternoon of environment wrangling, and an OS or dependency update can break it again.

The one-click way

3D Mate ships TripoSR as a one-click install. It packages the model, the Apple Silicon math, a proper background-cutout step, and export to GLB, USDZ, OBJ, STL, and PLY, and it runs fully offline with nothing to configure.

Download for macOS About TripoSR

Questions

Can TripoSR run on a Mac?
Yes. TripoSR is small and MIT-licensed, so it runs on Apple Silicon. The official code targets NVIDIA GPUs, so the mesh-extraction step needs a CPU substitute, or you can run it in one click inside 3D Mate.
How fast is TripoSR on Apple Silicon?
It produces a draft in roughly under a minute on a recent Apple Silicon Mac. The CPU marching-cubes path is slower. In 3D Mate it is under a minute on an M1 Pro.
Do I need an internet connection?
Only to download the model once. After that TripoSR runs offline. In 3D Mate the engine installs from inside the app and then runs with no network at all.