Features
- Generates 3D assets from a single image
- Uses a 4B-parameter flow-matching Transformer
- Produces watertight meshes with O-Voxel representation, including thin or open geometry
- Generates at up to 1536³ resolution, from about 3 seconds at 512³
- Outputs full PBR materials: base color, metallic, roughness and opacity
- Needs no per-asset optimization
- Paints existing 3D meshes from a reference image
- Exports to GLB and OBJ
- Open source under the MIT license, with a web demo and API
Pricing Details
💵 Pricing & Plans (February 2026 Status): Plan Price (Approx.) Key Features Open Source $0 (Free) MIT Licensed. Code and weights available on GitHub and Hugging Face. Hugging Face Space Free / Zero GPU Web-based demo for quick testing (subject to queue times). API (e.g., Fal.ai) Pay-per-gen Cloud-hosted version for high-speed, scalable batch generation. 🎁 How to Get Started: Visit the Microsoft TRELLIS.2 Hugging Face Space to upload a single image. Within seconds, you can interact with a 3D preview of the object. For developers, clone the GitHub repository and run the inference script; note that it requires an NVIDIA GPU with at least 24GB of VRAM (e.g., RTX 3090/4090 or A100/H100).