DeepMind, Google's artificial intelligence research lab, has introduced the Genie 2 AI model, capable of creating 3D game prototypes. The original Genie was released in February and could generate two-dimensional worlds.
For the new model to work, it is enough to “feed” it just one concept art, writes TechCrunch. The neural network itself decodes the image into components, animates the character and simulates lighting, physics, reflection, and “NPC” behavior.
The results are already cool: many examples of Genie 2's work look like scenes from AAA titles. Most likely, this is due to the fact that Google trained the model on data from popular video games, including No Man's Sky and Valheim. However, DeepMind, like many AI labs, does not reveal many details about its data acquisition methods for competitive reasons.
Currently, Genie 2 is able to create 3D worlds lasting up to a minute, while all the demonstrated examples from Google last 10-20 seconds. There were also artifacts and “hallucinations”.
It is not yet known when Genie 2 will be available to the public. The authors continue to work on the project, and hope to see it as a real tool for developers in the near future.
Creative people can have mixed feelings – especially those who work in the video game industry. A recent Wired investigation revealed that major players in the market, such as Activision Blizzard, which has laid off dozens of artists, have begun using AI to create content for their games.
However, Google is increasingly investing resources in its research, which promises to become “the next big thing” in the field of AI. Earlier, the search giant introduced a universal AI gamer for playing any game.