Text-to-motion is a thing now, demo on Hugging Face

This space uses T2M-GPT models based on Vector Quantised-Variational AutoEncoder (VQ-VAE) and Generative Pre-trained Transformer (GPT) for human motion generation from textural descriptions.