TRELLIS.2 is a cutting-edge open-source model and codebase for high-fidelity 3D asset generation from 2D images, developed to push forward the state of the art in image-to-3D generation. At its core is a novel sparse voxel structure called O-Voxel that jointly encodes both geometry and surface appearance, enabling reconstruction and generation of complex 3D shapes with arbitrary topology, open surfaces, and physically based rendering (PBR) textures. The system leverages a large 4-billion-parameter architecture combining sparse 3D variational autoencoders with flow-matching transformers to produce fully textured 3D models at resolutions up to 1536³ voxels. TRELLIS.2 emphasizes speed and compact latent representation, allowing bidirectional conversion between mesh formats and internal representations with minimal preprocessing and optimized performance on high-end GPUs.

Features

  • State-of-the-art image-to-3D generative model
  • O-Voxel structured latent representation
  • Full PBR texture support
  • High resolution output up to 1536³ voxels
  • Efficient bidirectional mesh-voxel conversion
  • Tools for dataset preparation and model inference

Project Samples

Project Activity

See All Activity >

License

MIT License

Follow TRELLIS.2

TRELLIS.2 Web Site

Other Useful Business Software
Build Data Resilience - Take the Assessment Today Icon
Build Data Resilience - Take the Assessment Today

Can you recover when it matters most? Take this quick assessment to identify gaps and build greater recovery confidence.

Is your recovery strategy as strong as you think? Take this quick self-assessment to check your recovery readiness and gain tailored insights. In only 2 minutes, you'll learn where you fall on the recovery readiness scale.
Take the Assessment
Rate This Project
Login To Rate This Project

User Reviews

Be the first to post a review of TRELLIS.2!

Additional Project Details

Operating Systems

Linux

Programming Language

Python

Related Categories

Python 3D Modeling Software, Python AI Models

Registered

2026-01-29