Google DeepMind's Genie 4 world model adds audio and real-time text steering

Key figure: Genie 3 announced Aug. 5, 2025, as a limited research preview. Source: Google DeepMind.[4]

What is Genie 4?

Google DeepMind’s new world model, announced Oct. 27. It generates interactive worlds in real time that users move through with keyboard controls and steer with text prompts. DeepMind says it is natively multimodal, generating music, ambient sound and dialogue with the world, and that Gemini 3.8 Flash runs in the loop to plan steps and check frames. All video and audio it makes will carry SynthID watermarks.[1]

What's new versus Genie 3?

DeepMind lists native audio, including dialogue (its first real-time world model to generate audio natively), text edits mid-simulation such as changing the weather, moving the camera or adding a person, and more realistic physics and people.[1] DeepMind announced Genie 3 on Aug. 5, 2025 as a limited research preview generating worlds at 24 frames per second.[4]

Who can use it, and when?

Only trusted testers, in a research preview. It is not yet available to Google AI Ultra subscribers or developers. DeepMind says it is “coming soon” to Ultra subscribers in Project Genie and to early access partners of the Gemini API and Google AI Studio, with a waitlist open.[1]

What is it for?

DeepMind cites training environments for robots and agents and new interactive experiences. Testers are trying it for film pre-visualization, game prototypes, scenes built from archival photos and enterprise scenario simulation. In one demo, Gemini 4 Argon broke a task into steps that Genie 4 simulated with Apptronik’s Apollo humanoid playing checkers.[1]

What don't we know yet?

No price has been announced.[1][2][3] DeepMind calls Genie 4 an early, experimental model: details and object permanence can drift in longer runs, and audio accuracy, legible text and complex physics are still research areas.[1] DeepMind’s Diego Rivas said the end-to-end product experience is not yet clear, TechCrunch reports.[2]