A short history
How TRELLIS.2 came to be.
From decades of 3D research to a one-click local installer — the lineage that made consumer-grade 3D generation possible.
From decades of 3D research to a one-click local installer — the lineage that made consumer-grade 3D generation possible.
TRELLIS.2 didn't appear in a vacuum. It builds on years of progress in implicit neural representations, diffusion models, and consumer GPU acceleration.
NeRF showed the world that a small neural network could store a fully view-consistent 3D scene. It was slow, but it was a proof: deep learning could model 3D.
Researchers showed that 2D diffusion models could supervise 3D generation. The results were dreamy, but the process took hours per asset — unusable for production.
3DGS replaced slow ray-marching with millions of tiny anisotropic gaussians, enabling photoreal scenes that render in milliseconds — opening the door for consumer 3D AI.
The original TRELLIS paper introduces SLAT — a sparse, voxel-anchored latent that can be decoded into meshes, gaussians, or radiance fields. A unified 3D foundation model is born.
TRELLIS.2 doubles the latent capacity, adds native text-to-3D, and ships with a friendly local UI. The model becomes small enough to run on an 8GB consumer GPU.
An open-source community project — the One-Click TRELLIS.2 Install — bundles the model, runtime, and UI into a single installer. Anyone with a GPU can now generate 3D assets locally in under a minute.
What once required a render farm now runs on a gaming laptop. What once took hours now takes seconds. And what once lived in research papers now lives on your hard drive — one click away.