Researchers from Google and Tel Aviv University have introduced GameNGen, an AI model capable of simulating the popular 1993 video game Doom in real time. By leveraging AI image generation techniques from Stable Diffusion, GameNGen serves as a limited game engine, potentially revolutionizing real-time video game synthesis. Instead of using traditional graphical techniques, future games could employ AI engines to generate graphics on the fly. GameNGen can generate new Doom gameplay frames at speeds above 20 frames per second using a single tensor processing unit.

In tests, human raters struggled to differentiate between actual gameplay footage and GameNGen outputs. This concept of "neural rendering" aligns with the ongoing research on AI video synthesis models like OpenAI's Sora. The research paper, titled "Diffusion Models Are Real-Time Game Engines, " explains GameNGen's process by using a modified version of Stable Diffusion 1. 4, an image synthesis diffusion model. However, using Stable Diffusion leads to graphical glitches and challenges with maintaining visual clarity and consistency over time.