World Simulation with Video Foundation Models for Physical AI

World Simulation is the core to scalable physical AI development. At CoRL 2025, NVIDIA announced major updates to Cosmos World Foundation Models (WFMs) that let developers generate diverse data for accelerating training physical AI models at scale using text, image and video prompts. 

Cosmos Predict 2.5 combines three WFMs into one, reducing complexity while enabling longer (up to 30s) video generation and multi-view outputs for richer simulations.

Cosmos Transfer 2.5 is 3.5x smaller yet faster and higher quality, producing photorealistic data from spatial inputs and ground truth simulations.

All Cosmos models are openly available under the NVIDIA Open Model License https://github.com/nvidia-cosmos