NVIDIA Cosmos3-Super is a large-scale foundation model designed for world simulation and high-fidelity video generation. It enables developers to create realistic video sequences that model physical interactions, object dynamics, and environmental behaviors from textual descriptions or input images. The model is distributed on Hugging Face with support for pip and Docker deployment, making it suitable for local inference and research experimentation in generative AI and simulation tasks.
In the Multimodal & vision space, Cosmos3 Super takes a focused approach. It focuses on generating realistic video simulations of physical world dynamics from text or image prompts. It is built as an open-source project for AI researchers and developers. Cosmos3 Super is open source under the Open Source license. It ships for the web, the command line, and API, and it can be self-hosted.
It is developed by NVIDIA (United States), and it first shipped in 2026. Development happens publicly on GitHub with 11.2k stars and 59 commits in the last 90 days. Key capabilities include Video Generation, World Simulation, and Multimodal Input.
Summary written by a language model from the project’s public pages.
What PulseGate has recorded for this listing
Closest matches by what these projects do