This repository implements T2M-GPT, a CVPR 2023 paper that generates human motion sequences from text prompts. It uses a VQ-VAE to learn discrete motion tokens from motion capture data, then employs a GPT-style transformer to autoregressively generate motion tokens from text-derived features. The model maps textual descriptions like ‘a man steps forward and does a handstand’ into corresponding skeletal animation sequences that can be rendered as SMPL meshes.