TripoSR generates a textured 3D mesh from a single source image, automatically creating geometry and textures not visible in the original image. The model was trained on a subset of the publicly available Objaverse research dataset, using 3D models licensed under the Creative Commons CC BY license.

In the technical report accompanying the release, Stability AI and Tripo AI compare TripoSR with other open-source 3D reconstruction models such as OpenLRM and One-2-3-45. TripoSR outperformed other tested models in terms of the accuracy of the generated models as well as processing time. In the demonstration video by Tripo AI, the results appear quite impressive, although the simpler models in the Stability AI blog video may be more typical for final results.

The initial user tests, such as this video from GamesFromScratch, show that the results for solid surface objects can be used as background models for games or AR applications, although the results for organic characters are... quite interesting.

The source code for TripoSR is available on GitHub under the MIT license, along with a list of dependencies. The model's weights are available on Hugging Face.
At default settings, the model requires around 6 GB of VRAM to process a single image, but according to Stability AI, it can operate on a PC without a GPU.