Google DeepMind details Genie 4’s capabilities and limits

Google DeepMind describes Genie 4 as its most realistic and interactive world model yet, and says it improves real-time interactive simulation.[3][1] Google says the model generates native audio while creating a world; its examples include ambient sounds, music and interactive dialogue.[1][2] The company says Gemini Omni anchors simulations in real-world appearance and behavior, while other Gemini models provide reasoning intended to support coherence and continuity.[1]

Users can steer scenes in real time with natural-language prompts, Google says.[2] Examples include moving the camera, changing the weather and introducing characters.[2] The Verge reports that Gemini 3.8 Flash helps guide Genie 4, and Ars Technica reports that it breaks goals into steps and evaluates output frames.[6][7] TechCrunch reports that Genie 4 can maintain a consistent scene for up to two minutes, twice Genie 3’s limit.[5] That duration is not a guarantee that details remain stable: TechCrunch reports that details and object consistency can drift as simulations run longer.[5]

The addition of native audio does not establish that generated sound will be accurate for every intended use. TechCrunch reports that Google has not yet resolved how to ensure audio accuracy or legible text inside Genie 4 worlds.[1][5] For teams assessing the system, this makes output review important, especially when a simulation depends on specific wording or sound cues. Treat generated scenes as material to inspect rather than assuming that audio and in-world text are dependable. Google also says Gemini models help keep simulations coherent and continuous, but TechCrunch’s reporting notes that scene detail and object consistency can still drift over time.[1][5]

Google says it has started rolling out Genie 4 to a limited group of trusted testers.[2] It says the model is coming to Google AI Ultra subscribers in Project Genie and to early-access partners using the Gemini API and AI Studio; DeepMind has opened a Project Genie waitlist.[2][1] Ars Technica reports that developer-access pricing had not been provided and that Google AI Ultra plans start at $100 per month.[7] Google says it is developing training environments for robots and agents, and its blog mentions filmmaking, journalism, gaming and manufacturing.[1][2] TechCrunch reports that Runway released its second world model, GWM Worlds 2, a couple of months before Genie 4; Diego Rivas says rapid ideation can help with ideation and storyboarding.[5]