TRELLIS 2

TRELLIS.2 is an open-source image-to-3D model generator, designed to produce high fidelity textured assets using native 3D VAEs. Its core feature makes use of native and compact structured latents, providing both high fidelity and compression capabilities. The method can handle complex structures, including open surfaces, non-manifold geometry, and enclosed interior structures, thus overcoming the limitations of iso-surface fields. TRELLIS.2 also has the ability to model arbitrary surface attributes such as Base Color, Roughness, Metallic, and Opacity, thereby facilitating Physically Based Rendering (PBR) and photorealistic relighting. This tool optimizes the pre- and post-processing of data for training and inference, allowing for quick conversions that are free from rendering and optimization. It utilizes 'O-Voxel', a novel 'field-free' sparse voxel structure designed to encode detailed geometry and complex appearance simultaneously. A Sparse Compression VAE component is deployed to compress voxel data efficiently, encapsulating fully textured 3D assets into a compact representation with minimal perceptual degradation. This enables efficient large-scale generative modeling. Its worth noting that TRELLIS.2 is purely a research project with Responsible AI considerations factored into all stages of its development.