🏃 PRISM: Text-to-Motion Generation

Generate human motion sequences from text prompts using PRISM — a flow-matching DiT transformer with per-joint latent decomposition for streaming motion synthesis.

The model outputs SMPL body parameters (22 joints) rendered as a 3D skeleton animation.

33 257
1 15
10 100
Examples
Motion Description Number of Frames Guidance Scale Inference Steps Seed