Researchers from Google, DeepMind, and Tel Aviv University have developed the first game engine that operates based on a neural model, enabling real-time interaction with complex environments.
GameNGen is a diffusion model trained on gameplay footage from Doom. It generates the next game frame based on previous frames and commands entered by the player.
Although GameNGen renders the game at a mere 20 frames per second, visually it is almost indistinguishable from the original. An experiment demonstrated that ten experts, shown 130 random short video clips, could barely tell the generated clips apart from actual Doom gameplay.
The developers believe that this technology could revolutionize the way video games are created. They argue that GameNGen serves as proof that a neural model can effectively run a complex game on existing hardware, which may make game development less costly and more accessible.
Scientists speculate that in the future, games could be developed and edited using text descriptions or image examples. However, questions regarding the creation of original content and the ethical implications of using existing games for training models remain unresolved.
Interestingly, enthusiasts have previously shown Doom running on a variety of unexpected platforms, ranging from intimate devices and John Deere tractors to biological systems, even employing gut bacteria.
Source: 404media
Comments (0)
There are no comments for now