• Picks
  • Discover
  • Alternatives
  • About

© 2026 Willet

  • Newsletter
  • About
  • FAQ
  • Terms
  • Privacy & Cookies
Join the Discord (opens in a new tab)
Microsoft avatar

TRELLIS.2

4B-parameter model that turns a single image into a textured 3D asset

  • AI media
  • AI models and runtimes
  • Design and creative
View on GitHub

What it is

A 3D generative model from Microsoft that takes one image and produces a mesh with PBR materials (base color, roughness, metallic, opacity), exportable to GLB. It uses a sparse voxel representation called O-Voxel that handles open surfaces, non-manifold geometry and internal structures, at resolutions from 512³ to 1536³. The repo ships inference code, training code, and shape-conditioned texture generation; it needs Linux and an NVIDIA GPU with 24GB or more. Researchers and 3D artists use it to generate game-ready or render-ready assets from reference pictures.

Images

Facts

Stars
12,340 GitHub stars.
12,340

Forks
1,491
Commits
11
Contributors
3
Language
Python
License
MIT
Last commit
Jun 5, 2026
On GitHub since
November 2025
Platforms
Linux, library
Self-hostable
No
Alternative to
Meshy, Tripo
Tags
  • open-weights
  • 3d-modeling
  • 3d-reconstruction
  • computer-vision

Last synced from GitHub on Oct 20, 2026

The 10 best overlooked open-source projects, every Monday.

Share and more

Comments

No comments yet. Start the conversation.

Similar projects

  • Linzhan Mou avatar

    UniMate

    Text-driven motion generation model for characters with any skeleton

    2.1k

What people are saying

Updated Oct 20, 2026
  • Hacker NewsTRELLIS.2: state-of-the-art large 3D generative model (4B)Dec 16, 2025 · 86 points · 17 comments (opens Hacker News in a new tab)