RigMo can auto-rig almost any 3D model for animation

Rigging a 3D model has long been one of the biggest headaches in computer graphics. Painting vertex skinning weights by hand is nobody's idea of fun, and one misplaced joint can make a character's shoulder twist into a melted noodle the moment it moves.

Researchers published a paper introducing RigMo, a unified framework that automatically extracts animatable skeletons and motions directly from raw 3D mesh sequences. Best of all, it works without a single human-provided skeleton annotation.

Learning bones straight from motion

Existing 4D animation pipelines usually treat auto-rigging and motion generation as two completely isolated problems. They tend to rely heavily on ground-truth bone structures created by artists, which severely limits how well they generalize across different character types.

RigMo bypasses that bottleneck by encoding per-vertex surface deformation into two latent spaces. The first is a rig latent that decodes into explicit Gaussian bones and skinning weights. The second is a motion latent that handles time-varying 3D rigid transformations.

By putting both pieces together, the model constructs an explicit, fully rigged animatable mesh. On top of that, the researchers introduced Motion-DiT, a diffusion transformer operating inside RigMo's latent space to handle downstream motion generation.

From walking bears to spinning desk fans

In benchmarks across DeformingThings4D, Objaverse-XL, and TrueBones, RigMo produced smooth, physically plausible rigs with strong category-level generalization.

It handles humans easily, but it also flexes its muscles on quadruped animals like bears, winged birds, and even rigid mechanical objects like rotating desk fans. Compared to prior approaches like Uni-Rig, RigMo produces significantly fewer skinning artifacts and keeps joints firmly in place instead of twisting them into knots.

The technical paper is already available on arXiv, and the authors state that code and datasets will be open-sourced soon.

Why it matters

For indie game developers and 3D animators, reliable auto-rigging could eliminate hours of tedious, repetitive manual labor. Turning raw geometry into animatable rigs with clean weights is the holy grail of asset creation pipelines. However, since only the paper is out right now, creators will have to wait for the code release before testing it in Blender or Unreal.

For the AI industry, RigMo solves a glaring practical problem. A lot of recent 4D video and mesh generators output dense, shifting vertex clouds. They look cool in demos, but studio artists cannot easily tweak or retarget them. Explicit skeletons and skinning weights are the universal language of production engines, and RigMo bridges that gap.

We think the smart move here is inferring structure from movement. Instead of guessing where joints belong on a static statue, watching how an object actually deforms makes bone placement much more reliable. The main limitation is that it currently requires raw mesh deformation sequences as input, but it points to a future where any generated 3D creature can be dropped straight into an interactive game world.