We propose a framework that utilizes bone density control to extract bone structure from garment motion sequences. A GRU network then leverages this bone structure to infer garment deformations from human motion, producing garment meshes that accurately follow body movements. Given a garment, we employ an example-based adaptive rigging method to extract virtual bones from its simulated mesh sequences. The density of virtual bones across different regions of the garment is controlled by the complexity of deformation. At runtime, a multi-layer GRU network takes the body’s motion sequence as input and predicts the transformations of the virtual bones, which are then blended to deform the garment mesh. Explicitly imposing constraints to maintain consistency in the position of the transformed virtual bones ensures the physical interpretability of the learned anchor transformations in space. Experiments demonstrate that our method outperforms state-of-the-art approaches in terms of prediction accuracy and visual quality.