The people behind it
Meet the inventors.
TRELLIS originated as a research project at Microsoft Research and Tsinghua University, led by Jianfeng Xiang and a team of 3D and generative modeling researchers.
TRELLIS originated as a research project at Microsoft Research and Tsinghua University, led by Jianfeng Xiang and a team of 3D and generative modeling researchers.
Jianfeng Xiang has spent his career at the intersection of computer graphics and machine learning, with a focus on 3D-aware generative models. His earlier work on neural implicit representations and 3D-aware GANs laid the conceptual groundwork for what would become TRELLIS.
The key insight behind TRELLIS — that geometry and appearance could share a single, sparse, voxel-anchored latent space — was a result of years of iteration. TRELLIS.2 extends that idea with larger latent capacity, native text conditioning, and a more efficient sparse transformer backbone.
The guiding principle behind TRELLIS.2: every creator — student, hobbyist, professional — should be able to generate high-quality 3D assets on their own hardware, without paying for cloud compute.
The original TRELLIS work was a collaboration between Microsoft Research Asia and Tsinghua University. TRELLIS.2 added contributors from the broader open-source community.
Microsoft Research Asia & Tsinghua University. Original SLAT representation, sparse transformer architecture, and training pipeline.
Open-source contributors who optimized the model for consumer GPUs, added native text-to-3D, and built the inference UI.
A community-driven packaging project that bundles TRELLIS.2 with its runtime and weights into a single installer for Windows.