TRELLIS: 3D generator — what it does and what you need to run it

TRELLIS is an open-source 3D generator created by Microsoft. It tackles the image-to-3D task, turning one or more images of an object into a textured 3D mesh. It is based on the research paper “Structured 3D Latents for Scalable and Versatile 3D Generation,” presented at CVPR'25.

What it does

As described by the authors, TRELLIS generates 3D models from input images. The result is saved as a textured polygon mesh in GLB format (file sample.glb). The repository is also tagged with the text-to-3D label, but the fact sheet does not detail this mode.

What you need to run it

The code is written in Python and distributed under the MIT license. The model weights are also available under the MIT license. The largest weight file takes up 1.1 GB, and the full set of weights is 3.1 GB. To run it, as stated by the developers, the CUDA platform is required. The documentation notes that CUDA Toolkit is needed to compile some submodules; the code was tested with versions 11.8 and 12.2. Running locally without an NVIDIA GPU is not supported, judging by these requirements. No cloud service version is listed in the fact sheet — this is a local tool.

Who it is for

The tool will be useful for researchers and developers experimenting with generative 3D who have access to a machine with an NVIDIA GPU. The open MIT license allows embedding TRELLIS into your own projects and modifying the code. Those working on integrated graphics or AMD GPUs will not find the tool suitable due to its tight CUDA dependency. It is also not designed for users expecting a ready-made cloud service: the authors do not mention such an option, and all interaction goes through code. TRELLIS occupies the niche of a research-grade generator that offers full control over the process but requires technical expertise and compatible hardware.