hugging-apps/prism-text-to-motion

PRISM: Streaming Human Motion Generation with Per-Joint Latent Decomposition

This Space demonstrates PRISM, a text-to-motion generation model that produces SMPL body motion sequences from natural language prompts.

How it works

1. 1.Enter a text prompt describing a human motion
2. 2.The model generates a motion sequence using a flow-matching DiT transformer with a causal spatio-temporal Motion VAE
3. 3.The output is rendered as a 3D skeleton animation

Model

• —Model: ZeyuLing/PRISM-TP2M-1.4B (~1.4B parameters)
• —Architecture: Flow-matching DiT transformer with per-joint latent decomposition
• —Text encoder: UMT5 (T5-style)
• —Output: SMPL body parameters (22 joints, rotation_6d, 30 fps)

Citation

@article{ling2026prism,
  title={PRISM: Streaming Human Motion Generation with Per-Joint Latent Decomposition},
  author={Ling, Zeyu and Shuai, Qing and Zhang, Teng and Li, Shiyang and Han, Bo and Zou, Changqing},
  journal={arXiv preprint arXiv:2603.08590},
  year={2026}
}