New AI model GameGen-O creates open-world video game simulations

These are not fully playable games, but rather a tool that developers can use to quickly create prototypes and test different game elements without having to build them from scratch, it seems.

The demo videos clearly show scenes from copyrighted games, with well-known franchises easily recognizable. This will likely spark further discussions about copyright issues.

https://www.youtube.com/watch?v=3WFdYFmDyuo

To create GameGen-O, the researchers first built a large dataset called "OGameData," which includes processed data from over 100 modern open-world games. They then trained the model in two stages: First, it learned to generate various game content based on OGameData. Next, they added an "InstructNet" component to enable interactive control of the generated content.

Advancing generative AI for game development

The researchers see GameGen-O as a significant step forward in using AI to generate open-world video games. It demonstrates the potential to serve as an alternative to traditional rendering techniques by combining "creative generation with interactive capabilities."

According to the team, GameGen-O could be a valuable resource for researchers and developers. It allows for exploring various applications such as developing AI for video games, interactive control, and immersive virtual environments.

https://www.youtube.com/watch?v=Mhg3f0Lm1-I

Recently, researchers from Google Research, Google DeepMind, and Tel Aviv University demonstrated GameNGen, an AI system that can fully simulate and play back the classic game DOOM in real time at over 20 frames per second.