What it does
From a prompt to a production-ready asset.
TRELLIS.2 is a unified 3D generative model. It takes images or text and produces clean, textured 3D assets ready for any pipeline — all running locally on your GPU.
TRELLIS.2 is a unified 3D generative model. It takes images or text and produces clean, textured 3D assets ready for any pipeline — all running locally on your GPU.
Feed TRELLIS.2 a single image — a product photo, a concept render, a sketch, even a screenshot — and it will reconstruct a full 3D model with topology and texture in seconds.
Internally, the input is encoded into the Structured 3D Latent space. From there, you can decode a watertight mesh, photorealistic 3D gaussians, or an editable radiance field.
One unified pipeline, every output format your tools need.
TRELLIS.2 adds native text conditioning. Type a prompt — "a weathered viking helmet with bronze accents" — and the model produces a 3D asset that matches.
No intermediate image generation step. No SDS loop that runs for hours. Just direct, end-to-end 3D synthesis on your local hardware.
Clean, manifold triangle meshes with UV-mapped PBR textures. Export to GLB, OBJ, or FBX.
Millions of anisotropic gaussian splats for photoreal real-time rendering. PLY export.
Editable NeRF-style radiance fields for high-fidelity novel-view synthesis.
Auto-generated albedo, normal, roughness, and metallic maps — ready for any engine.
Generate multiple variations of the same prompt with reproducible seeds.
Lock geometry, regenerate texture. Lock texture, regenerate geometry. Iterate fast.
Built-in remeshing and decimation. Choose your target polygon count on export.
The One-Click Installer ships with a browser UI — drag, drop, generate, download.
8GB VRAM minimum. Windows 10/11 with an NVIDIA GPU. One-Click TRELLIS.2 Install handles drivers, CUDA, dependencies, and weights — automatically.