Term
Microsoft TRELLIS.2
Microsoft TRELLIS.2 is the strongest open image-to-3D model (4B parameters, MIT license). It generates high-resolution assets with native PBR materials in seconds and supports Gaussian Splatting; weights are freely available on Hugging Face.
Microsoft TRELLIS.2 — explained in more detail
Microsoft TRELLIS.2 is a large, open 3D generative model for high-fidelity image-to-3D generation. With 4 billion parameters it converts images into high-resolution 3D assets in seconds, with full PBR materials and complex topology. Technically it uses a novel field-free sparse voxel structure (O-Voxel) and a large flow-matching transformer; its 3D VAEs operate at 16x spatial compression and support resolutions from 512 to 1536 cubed.
The decisive difference from Tripo, Rodin and Meshy is its openness: model and code are released under the MIT License, so they can be used for any purpose including commercial projects. The pretrained weights (model TRELLIS.2-4B) are hosted on Hugging Face, and the GitHub project includes code, tooling and examples. This makes TRELLIS.2 the strongest open 3D model currently available, with native PBR materials and support for Gaussian Splatting.
Example / In practice
A studio with data-privacy requirements can run TRELLIS.2 locally on its own hardware to generate 3D assets from product photos without sending images to an external service — and, thanks to the MIT license, may use the outputs commercially without restriction.
Distinction from similar terms
TRELLIS.2 shares the 3D-generation use case with Tripo AI, Hyper3D Rodin and Meshy, but differs fundamentally through its open weights (open-weight pill). While the commercial platforms run as hosted services and lead in some quality or convenience dimensions, TRELLIS.2 is self-hostable and customisable, making it the reference in the open segment.