Google DeepMind introduces Genie 4, a research-preview AI model designed to generate interactive, photorealistic virtual worlds. Across outlets, Genie 4 is described as being able to produce native audio—such as music and ambient sounds—and support dialogue within simulations. Several reports also highlight that users can steer a running scene using natural-language prompts, including changing elements like the weather, camera angle, or adding people and objects mid-simulation.
Outlets describe Genie 4 as an upgrade from Genie 3, including improved realism, better physics, and more human-like characters. Multiple sources say the model can maintain a consistent scene for up to about two minutes, with Ars Technica and Quartz framing this as double the earlier limit. Coverage notes that Genie 4 is being rolled out to trusted testers first, with a broader release for Google AI Ultra subscribers planned in coming weeks.
Some differences in emphasis appear in how outlets describe the “steering” interaction (text prompts versus “real-time steering” with natural-language prompt), and which technical details they foreground, such as mention of Gemini components. The underlying capabilities—audio generation, real-time editing, and longer scene consistency—are consistently reported.
