Vi3W
Skip to content

Guides

TripoSR explained: what it is, how to run it locally and when to use it

TripoSR explained: the MIT-licensed image-to-3D model from Tripo and Stability AI, VRAM needs, local install, ComfyUI, and when to use it.

2 min read

A photo card of a toy car with a motion streak turning it into a solid 3D toy car
Illustration. TripoSR's selling point is speed on modest hardware.

Quick answer

What is TripoSR?

TripoSR is an open-source model that reconstructs a 3D mesh from a single image, released in 2024 by Tripo (VAST) and Stability AI under the MIT licence, covering code, weights and demo. Its default settings use about 6 GB of VRAM per image, which makes it one of the lightest open image-to-3D models to run locally.

TripoSR in brief

  • What: feed-forward 3D reconstruction from one image.
  • Who: Tripo (VAST) with Stability AI.
  • When: repository created February 2024.
  • Licence: MIT for source, pretrained models and demo.
  • Hardware: about 6 GB VRAM by default for one image; CUDA recommended.
  • Repo: VAST-AI-Research/TripoSR.

"Feed-forward" means one pass through the network, not a long diffusion loop, which is why it's fast.

When to use TripoSR

  • Quick tests: does this image convert at all?
  • Blockouts and placeholders for games.
  • Batch-converting many simple objects.
  • Small GPUs where TRELLIS (16 GB) or TRELLIS 2 (24 GB) won't fit.

When not to: hero assets, detailed characters, or anything where texture quality matters. Newer models are much stronger there. See TripoSR vs TRELLIS.

Run TripoSR locally

  1. Clone the repo.
  2. Install PyTorch matching your CUDA version (the README stresses matching CUDA major versions).
  3. pip install -r requirements.txt.
  4. Run the provided script with your image path; results save to output/. Several images can be passed at once.
  5. Optional: the Gradio demo.

Known error: torchmcubes was not compiled with CUDA support means the CUDA versions don't match; reinstall with matching versions per the README.

A toy car photo turning into a 3D car
Illustration. One image in, one quick mesh out.

TripoSR in ComfyUI

The ComfyUI 3D Pack lists TripoSR among its supported models. See ComfyUI 3D Pack.

TripoSR vs Tripo's commercial models

TripoSR is a research model. Tripo's commercial v3.1, P1 and P2 are different, closed and far more capable. The team's later open shape model is TripoSG (MIT, 8 GB+). See Tripo AI open-source alternative.

Skip the install

If you need better-than-draft results, use a stronger model. Vi3W runs TRELLIS (free plan), TRELLIS 2, Tripo v3.1, P1 and P2 and 20 more models in the browser, with retopology and GLB, FBX or OBJ export. Try Vi3W.

Frequently asked questions

triposr local

Clone the repo, install PyTorch for your CUDA version and the requirements, then run the script on an image. Output is saved to the output folder. A Gradio demo is included.

triposr for blender

Export the mesh, import it into Blender, and expect simple shapes and basic colour. Use it for blockouts and drafts, then remodel or retopologise.

triposr text to 3d

TripoSR is image-only. For text, generate an image first with any image model, then feed it to TripoSR.

Last reviewed