Google’s new AI model creates video game worlds in real time

Google Deepmind publishes a new version of its AI “World” model, called Genie 3, capable of generating 3D environments with which users and agents of AI can interact in real time. The company also promises that users will be able to interact with worlds much longer than before and that the model will really remember where things are turned off.

Global models are a type of AI system that can simulate environments for education, entertainment, or to help train robots or AI agents. With the world models, you give them an prompt and they generate a space in which you can move as you would in a video game, but instead that the world is handcrafted with 3D active ingredients, everything is generated with AI. It is an area in which Google puts a lot of effort; The company showed Genie 2 in Decemberwho could create interactive worlds depending on an image, and it builds A world model team led by an old co-directive of OPENAI Sorai Video Generation tool.

But the models currently have a lot of drawbacks. The Genie 2 worlds were only players up to a minute, for example. I recently tried the “interactive video” of a company Supported by the co -founder of PixarAnd it was like walking in a vague version of Google Street View where things turned and changed in a way I did not expect while looking around.

Genie 3 seems to be a notable step. Users will be able to generate worlds with an prompt that supports a few “a few” minutes of continuous interaction, which is up from 10 to 20 seconds of possible interaction with Genie 2, According to a blog article. Google says that Genie 3 can keep the spaces in visual memory for about a minute, which means that if you turn away from something in a world and then come back, things like painting on a wall or writing on a blackboard will be in the same place. The worlds will also have a 720p resolution and will operate at 24 frames per second.

Deepmind also adds what he calls “rapid global events” in Genie 3. Using an prompt, you can do things like changing weather conditions in a world or adding new characters.

However, this is probably not a model that you can try for yourself. It is launched as “a limited search overview” which will be available for “a small cohort of academics and creators” so that its developers can better understand the risks and how to mitigate them appropriately, according to Google. There are also many restrictions, such as ways limited by users, users can interact with the generated worlds and this readable text is “often generated only when provided in the description of the world of entries”. Google says it is “exploring” how to bring Genie 3 to “additional testers” on the line.