TRELLIS

TRELLIS is a local AI model for generating 3D assets from images or text prompts, aimed at 3D artists and researchers exploring asset creation. It can produce meshes, radiance fields and 3D Gaussians from the same underlying representation, so you can choose an output suited to your rendering or editing work.

Its Structured Latent (SLAT) representation captures both geometry and appearance. That shared representation supports several output formats rather than tying generation to a single kind of 3D asset. The project also shows GLB exports with appearance baked onto meshes.

Editing goes beyond whole-object regeneration. You can create variants of an existing asset or change a targeted region using text or image prompts, such as replacing a robot's legs or adding weapons. The model works best with artistic assets; its ability to generate photorealistic real-world objects is limited. The project recommends image prompts for better results.

The Python code runs on your own hardware and has been tested on Linux. It requires an NVIDIA GPU with at least 16GB of VRAM. Pretrained models are available through Hugging Face and can load from local files, while a Gradio demo provides a browser interface to the locally running model.

The repository is open source under the MIT license. The project website separately restricts its page materials to academic and research use. Researchers can also use the training and fine-tuning code and a curated dataset drawn from Objaverse(XL), ABO, 3D-FUTURE, HSSD and Toys4k.