Jay Zhangjie Wu

OmniDreams Runs Generative AV Simulation in Real Time

A Cosmos-derived causal diffusion model uses simulator state, action cues, and a streaming KV cache to render 704×1280 rollouts at 68–105 FPS.