FloodDiffusion: Text → Human Motion

Motion streams as you type via the paper's diffusion-forcing schedule — committed frames render progressively from left to right.
Powered by ShandaAI/FloodDiffusionTiny.

Motion description

Length (latent tokens ≈ frames / 4)

10 120

Smoothing α (0 = max smooth, 1 = none)

0 1

Start typing — motion streams in here

Examples

FloodDiffusion – Cai et al., 2025 (arXiv:2512.03520). Apache 2.0 license.