GameNGen
A neural-network-based game engine world model developed by Google Research, capable of simulating complex game environments in real time. It is considered a representative work in the field of video-generative world models.
Core Facts
Timeline (last 90 days)
A world model is a type of model that learns the dynamic laws of an environment and can predict state evolution; typical examples include Ha & Schmidhuber's World Models, the Dreamer series, and generative environment models such as Genie and GameNGen.
GameNGen is capable of real-time neural rendering of the game Doom.
GameNGen renders Doom footage at approximately 20 frames per second on a single TPU
Representative works in world models include DeepMind's Genie, Google's GameNGen, and Tesla's internal FSD world model used for autonomous driving.
The Decart team's GameNGen was the first to demonstrate the ability to generate playable DOOM visuals in real time using a diffusion model.
All Facts (5)
A world model is a type of model that learns the dynamic laws of an environment and can predict state evolution; typical examples include Ha & Schmidhuber's World Models, the Dreamer series, and generative environment models such as Genie and GameNGen.
65%UnverifiedGameNGen is capable of real-time neural rendering of the game Doom.
50%UnverifiedGameNGen renders Doom footage at approximately 20 frames per second on a single TPU
50%UnverifiedRepresentative works in world models include DeepMind's Genie, Google's GameNGen, and Tesla's internal FSD world model used for autonomous driving.
50%UnverifiedThe Decart team's GameNGen was the first to demonstrate the ability to generate playable DOOM visuals in real time using a diffusion model.
50%