🚀 Introducing Cosmos-Predict2 — NVIDIA’s new open-video model for Physical AI!
This product is a vital part of the World Foundation Models (WFMs) ecosystem and is specifically designed for tasks in the field of Physical AI. The model is capable of predicting the future state of the visual environment using text instructions and video files. Cosmos-Predict2 aims to accelerate the development of systems that understand physics, the environment, and motion — whether autonomous vehicles, robots, or other complex devices. Everything looks very exciting.
This is the most advanced generation of models within the Cosmos system, significantly surpassing previous versions like Predict1:
🎯 improved video quality
🧠 better alignment with textual descriptions
🎥 more realistic motion dynamics
Additionally, Cosmos-Predict2 demonstrates superior results compared to other open foundational video models.
What’s included:
▪ model weights
▪ a complete set of tools for inference and training, along with detailed tutorials
If you’re interested — stay tuned for updates!