SuperAPI AI Directory
Partner with us中文AI newsFavorites
DirectoryNews and tutorials

TRELLIS.2 - Microsoft's open-source 4-billion-parameter 3D generation model

Author: AI小集 · Original source: AI工具集

What is TRELLIS.2

TRELLIS.2 is Microsoft's open-source 4-billion-parameter 3D generation model, designed for efficiently generating from a single imagehigh-fidelity 3D asset design. The core innovation lies in its O-Voxel sparse voxel architecture, which can handle complex topologies (such as open surfaces and non-manifold geometry) and supports full PBR material modeling. Generating a 512³ resolution model on an NVIDIA H100 takes only 3 seconds, with support for up to 1536³ ultra-high-definition rendering. Compared with traditional methods, it significantly improves efficiency through 16x spatial downsampling, while also providing GLB format export functionality, making it suitable for fields such as gaming, film and television, and industrial design.

TRELLIS.2

Main features of TRELLIS.2

  • Image-to-3D generation: It can quickly generate high-quality 3D models from input images, making it suitable for scenarios where 3D assets are created from 2D images.
  • Text-to-3D generation: It supports generating 3D models based on text descriptions, and users can generate the desired 3D objects through text instructions.
  • Complex topology handling: It can handle complex 3D structures, including open surfaces, non-manifold geometry, and internal closed structures, generating 3D models with complex shapes.
  • Material and texture generation: The generated 3D models support multiple material properties (such as base color, roughness, metallic, opacity, etc.), and support transparency and realistic rendering.
  • Multi-resolution support: It supports 3D model generation at multiple resolutions from 512³ to 1536³, and users can choose the appropriate resolution according to their needs.
  • Fast rendering and export: The generated 3D models can be used directly for rendering and support export to standard 3D file formats such as GLB, making them easy to use in other software.
  • Model editing and variant generation: Users can edit the generated 3D models and generate different variants of the same object to meet diverse design needs.

Technical principles of TRELLIS.2

  • O-Voxel representation: It adopts the "O-Voxel" sparse voxel structure, an innovative 3D representation method that can efficiently handle complex topologies and sharp features, breaking through the limitations of traditional isosurface fields and supporting open surfaces, non-manifold geometry, and internal closed structures.
  • Sparse 3D VAE encoding: Through a sparse 3D variational autoencoder (VAE), 3D assets are encoded into a compact latent space, achieving efficient compression and generation, supporting 16x spatial downsampling, and significantly improving processing efficiency.
  • Full texture modeling: It not only handles base color but can also simulate PBR material properties such as roughness, metallic, and opacity, support transparency and realistic rendering, and generate 3D models with rich texture details.
  • Minimalist data processing pipeline: Data processing requires no complex rendering or optimization steps, and the conversion between textured meshes and O-Voxels can be completed in a short time, greatly simplifying the processing pipeline and improving generation speed.
  • Efficient rendering and post-processing: By combining CUDA-accelerated mesh processing tools (such as CuMesh) and efficient rendering techniques, it quickly generates high-quality 3D models and supports real-time rendering and post-processing operations.
  • Large-scale pretraining: Pretrained on large-scale datasets, the model has learned rich 3D shape and texture features and can generate diverse 3D assets based on input images or text.

TRELLIS.2 project address

  • Project Website：https://microsoft.github.io/TRELLIS.2/
  • Github Repository：https://github.com/microsoft/TRELLIS.2
  • HuggingFace Model Library：https://huggingface.co/microsoft/TRELLIS.2-4B
  • arXiv Technical Paper：https://arxiv.org/pdf/2512.14692
  • Online Demo：https://huggingface.co/spaces/microsoft/TRELLIS.2

Application Scenarios of TRELLIS.2

  • Game Development: Quickly generate high-quality 3D game assets, such as characters, props, scenes, etc., improving development efficiency and reducing production costs.
  • Film and Television Production: Used to create realistic 3D special effects and virtual scenes, supporting complex materials and textures, meeting the film and television industry's high requirements for visual effects.
  • Virtual Reality (VR) and Augmented Reality (AR): Generate immersive 3D environments and objects, enhancing user experience, suitable for VR/AR application development.
  • Architectural Design and Visualization: Quickly generate 3D visualization of architectural models and interior design, helping designers and clients better understand and evaluate design proposals.
  • Industrial Design and Product Development: Create 3D models of product prototypes, accelerating the design iteration process, facilitating virtual presentations and functional testing.
  • Education and Training: As a teaching tool, it helps students and trainees better understand and master the relevant knowledge and skills of 3D modeling, rendering, and design.