UniRig

UniRig is an open-source model that rigs 3D models automatically: it predicts a skeleton and then the skinning weights that tie the mesh to those bones. It comes from Tsinghua University and Tripo (VAST AI Research), and the paper was published at SIGGRAPH 2025.

According to the project's README, rigging runs in two stages. A GPT-style transformer predicts a valid bone hierarchy one token at a time, then a second network works out how much each bone moves each vertex. The same model is meant to work across people, animals and objects; the README shows an animated dragon and rabbit, and its sample input is a giraffe. It accepts OBJ, FBX, GLB and VRM files and outputs rigged FBX or GLB.

The code is MIT-licensed and free. The README says generation needs a CUDA GPU with at least 8 GB of VRAM, and installation involves several CUDA-specific packages. The released checkpoint is trained on the Articulation-XL2.0 dataset; the full model trained on Rig-XL and VRoid from the paper is still listed as a future release, as is bone-attribute prediction for physics.

For game developers, UniRig fills the step between generating a mesh (with tools such as Tripo or TRELLIS) and animating it. The team now points users to SkinTokens, a successor that combines both stages. AIGamer has not tested UniRig.