AI · AI model
microsoft/TRELLIS.2-4B
Open-source 4B flow-matching model generating textured 3D assets from single images.
TRELLIS.2-4B is an open-source image-to-3D foundation model developed by Microsoft that creates 3D meshes with physically based rendering (PBR) materials from a single input image. It utilizes a sparse O-Voxel representation to support complex topologies, open surfaces, and translucent materials without relying on iso-surface fields. Output meshes can be exported to standard formats like GLB for use in 3D production pipelines.
What it's for
- Generate textured 3D assets from concept art or reference photos
- Produce game-ready GLB models with PBR material channels
- Texture existing 3D meshes based on reference imagery
- Prototype 3D props quickly for visual effects or CGI
Features
- O-Voxel sparse representation handling non-manifold geometry and open surfaces
- Flow-matching transformer architecture with 4 billion parameters
- Generates PBR materials including support for transparency and translucency
- High-resolution voxel grid outputs up to 1536³
- Direct GLB export with texture decimation and remeshing controls
- Fast inference completing 512³ models in roughly 3 seconds on an H100 GPU
Pricing
Open source. Free (MIT License)
Details
- Type
- AI model
- Category
- AI
- Topic
- 3D Generation
- Open source
- Yes
- Platforms
- Linux, Code library, Self-hosted
- For
- Game developers, 3D artists, VFX technical directors, and AI researchers.
- Alternative to
- Tripo3D, Meshy, Luma Genie, CSM 3D
- Tags
- image to 3d, mesh generation, pbr, huggingface, open weights, microsoft
Why it was saved
- @evolving.ai: Generates production-ready 3D models with high fidelity and PBR textures quickly for game development, CGI, and 3D design workflows.
Compare
Related
Turn text prompts and 2D images into textured, printable 3D models.
Turn freehand sketches into textured 3D models inside a Three.js scene
Rebuilds reference images as procedural, animation-ready Three.js models in code.
AI model that generates textured 3D assets from single images
AI generator converting text and 2D images into 3D models.
Spatial AI models that turn images, text, and video into 3D worlds
Updated