Rong Xie

Directly conditioning video diffusion on audio wastes massive capacity on static background and identity pixels—routing control transitively through a causal motion latent distilled under a single frozen video teacher achieves real-time streaming at 15.4 FPS with zero fidelity loss.

Imagine turning any single-view video into a fully explorable, dynamic 4D scene – this paper gets us closer with a novel diffusion-based approach and a new large-scale dataset.