Skin Tokens: A Learned Compact Representation for Unified Autoregressive Rigging

SkinTokens uses an FSQ-CVAE to map high-dimensional skinning weights into a compact, discrete token representation. This approach enables TokenRig, a unified autoregressive framework that predicts skeletal parameters and skin deformation tokens in a single sequence to capture dependencies between skeleton structure and vertex weights.