Google DeepMind’s new AI model Genie 3 can create 3D interactive worlds in real time

Google DeepMind has unveiled a new AI model called Genie 3, that can generate 3D worlds while being interactive. Users just need to enter a text prompt describing the environment that the model will then simulate in real time at 24 frames per second, maintaining consistency at 720p for a few minutes.

Last year, the team had released their first “world models,” Genie 1 and Genie 3. Additionally, advanced AI video generation models like Veo 2 and Veo 3 also demonstrate an understanding of the physical world.

A blog posted with the release said that world models, which are able to understand environments and then re-create them help agents to predict which the environment changes and how their actions can affect it. 

“World models are also a key stepping stone on the path to AGI, since they make it possible to train AI agents in an unlimited curriculum of rich simulation environments,” it noted. 

The team has said that compared to Genie 2’s interactive window which lasted between 10 to 20 seconds, Genie 3 offers a “few minutes” of interaction. The AI model is also able to be more consistent with visuals so if a user moves away from a location and returns to it later, the spot looks the same.

However, Genie 3 isn’t available for public preview yet and will be rolled out to a select group of creators for testing.