GameNGen, a generative AI model from Google, simulates the video game Doom, producing frames at 20 FPS with human evaluations showing slight preference over actual gameplay.
Utilizing the open-source Stable Diffusion model, GameNGen generates gameplay frames based on prior frames and action inputs, allowing for intricate simulation of game states.
Google's evaluation revealed that judges preferred clips from the simulated GameNGen 40% of the time, demonstrating the effectiveness of the generative model.
This paradigm could revolutionize game development, making it cheaper and more accessible, allowing for game modifications and new content creation based solely on text descriptions or images.
Collection
[
|
...
]