About the Project
What is TRELLIS.2?
A next-generation 3D generative model that produces meshes, gaussians, and radiance fields from a single image or text prompt — and runs locally with one click.
A next-generation 3D generative model that produces meshes, gaussians, and radiance fields from a single image or text prompt — and runs locally with one click.
For years, generating a usable 3D model from an idea was either too slow (hand-modeling), too lossy (photogrammetry), or too inaccessible (research-only pipelines that required a graduate degree to install).
TRELLIS.2 was built to change that. It is a unified 3D foundation model — one network, many output formats — designed from day one to be small enough to run on a single consumer GPU, and friendly enough that anyone with a laptop can use it.
And because the entire pipeline runs on your machine, your assets, prompts, and reference images never leave your computer.
TRELLIS.2 generates a single latent — and from that same latent, it can decode a mesh, a 3D gaussian splat, or a radiance field. Pick the format that fits your pipeline.
Structured 3D Latents (SLAT) capture both geometry and appearance in a sparse voxel-anchored representation — one space, every output format.
No cloud APIs, no recurring fees, no privacy compromises. The model and weights live on your disk and run on your GPU.
The One-Click Installer takes care of CUDA, Python, drivers, and weights. You just click "Install" and start generating.