A team from Google and Tel Aviv University has introduced GameNGen, a generative AI game engine that simulates Doom at over 20 frames per second using innovative reinforcement and diffusion models.
GameNGen differs from traditional game engines by generating the entire game state and frames dynamically based on player actions and previous frames, avoiding reliance on extensive game footage.
The training phase involved creating a reinforcement learning agent to play Doom, whose data was used alongside a custom diffusion model to produce game visuals.
While GameNGen achieves around 20 FPS, it showcases significant potential with claims of reaching nearly 50 FPS, albeit with reduced visual quality when optimizing performance.
[
Collection
]
[
|
...
]