SkinTokens predicts full rigs for any 3D character, from humans to crabs

SkinTokens is a new AI method that takes a static 3D mesh and automatically infers both the internal skeleton (rig) and skin weights needed for animation. Once processed, characters can be dropped into game engines or animation tools and moved naturally without manual rigging.

Examples show SkinTokens handling standard humanoid models and more unusual forms like birds, quadrupeds, snakes, sharks, cartoon capybaras, crabs, beetles, and scorpions. In each case, it predicts plausible joint structures and deformation behavior so limbs bend and bodies flex correctly during animation.

Benchmark comparisons put SkinTokens ahead of previous rig predictors on average accuracy metrics, making it the most reliable automatic rigging approach tested so far. This can drastically cut setup time for 3D artists and indie game devs who work with large libraries of character meshes.

The team has published a technical paper and a GitHub repository. Code and training details are expected later, which will determine how easily creators can integrate SkinTokens into Blender, Unreal, or Unity pipelines.