Google DeepMind’s new Genie 2 is a large foundation world model that generates interactive 3D worlds that are being likened to video games. “Games play a key role in the world of artificial intelligence research,” says Google DeepMind, noting “their engaging nature, challenges and measurable progress make them ideal environments to safely test and advance AI capabilities.” Based on a simple prompt image, Genie 2 is capable of producing “an endless variety of action-controllable, playable 3D environments” — suitable for training and evaluating embodied agents — that can be played by a human or AI agent using keyboard and mouse inputs.

Genie 2 “can generate an interactive, real-time scene from a single image and text description (e.g. ‘A cute humanoid robot in the woods’),” making it “similar to models under development by Fei-Fei Li’s company, World Labs, and Israeli startup Decart,” writes TechCrunch.

Genie 2 builds on the Genie 1 model DeepMind unveiled in February. “In Genie 1, we introduced an approach for generating a diverse array of 2D worlds. Today we introduce Genie 2, which represents a significant leap forward in generality. Genie 2 can generate a vast diversity of rich 3D worlds,” DeepMind writes in a blog post.

Examples provided by Google DeepMind show people interacting with Genie 2 worlds generated from a single image created by Imagen 3, DeepMind’s text-to-image model. “This means anyone can describe a world they want in text, select their favorite rendering of that idea, and then step into and interact with that newly created world (or have an AI agent be trained or evaluated in it),” DeepMind writes.

Genie 2 can generate consistent worlds for up to a minute, though most of the examples shown in the blog post run 10 to 20 seconds.

SiliconANGLE calls Genie 2 a “far-reaching development that could pave the way for developers to create expansive and playable 3D game worlds using a simple text prompt.” For the moment, however, the worlds “suffer from problems with stability — quite simply, they’re not that stable, and the model begins to lose coherency after around 20 seconds or so.”

But AI technology is progressing quickly, and as far as Genie 2 is concerned, TechCrunch says “creatives may have mixed feelings — particularly those in the video game industry,” citing a recent Wired article discussing how companies like Activision Blizzard have laid off workers due to AI use.

DeepMind also introduced a weather predictive model, GenCast that was trained on data from the European Center for Medium-Range Weather Forecasts, SiliconANGLE reports, adding that it can generate 15-day weather forecasts in only eight minutes with a single TPU v5 tensor processing unit, when by comparison “the ECMWF system requires an enormous supercomputer with tens of thousands of processors to generate the same, physics-based forecast.”