NVIDIA also announced Cosmos 3 Edge, a new 4 billion parameter model built to run locally in real time along with cosmos dreams.
NVIDIA research and engineering leaders unveiled a series of new AI driven graphics and simulation tools during a keynote at the “SIGGRAPH” conference in Los Angeles, outlining how neural rendering and world models are reshaping creative and industrial applications for game studios, filmmakers, and robotics developers alike.
NVIDIA founder and CEO Jensen Huang said creators need tools capable of expanding imagination while offering the precision to realize ideas exactly as intended.
Edward Liu, NVIDIA’s director of applied deep learning research, opened the technical presentation by showcasing 3D guided neural rendering, addressing challenges like preserving artistic intent, maintaining visual consistency across frames, and rendering high resolution content in real time.
He highlighted the NVIDIA Earth 2 model family, which allows scientists to boost simulation resolution using AI trained on simulation data while matching or exceeding the accuracy of traditional methods.
Ming-Yu Liu, vice president of NVIDIA’s Cosmos Lab, detailed updates to the Cosmos world foundation model platform, which is designed to accelerate development of physical AI systems. He explained that Cosmos was trained across diverse physical tasks using a single model architecture.
NVIDIA also announced Cosmos 3 Edge, a new 4 billion parameter model built to run locally in real time, along with cosmos dreams, a set of closed loop simulators that can generate full virtual environments from a single image.
Beyond the keynote, several major creative software platforms, including Adobe, Blender, and Unreal Engine, announced new model context protocol connections, allowing AI agents to work directly within design and production tools.
