Google DeepMind Unveils Genie 4 With Native Audio and Dynamic Steering

Real-time steering and native sound

Genie 4 allows users to adjust continuous simulations dynamically while generating synchronized environmental audio.

Interactive world models have historically struggled with two major bottlenecks: immediate responsiveness to prompt changes and consistent sound. With Genie 4, Google DeepMind aims to bridge that gap by enabling on-the-fly steering. Rather than rendering a fixed scenario from a single initial prompt, users can type instructions mid-simulation to adjust camera perspectives, alter character actions, or introduce sudden scene events.

Audio is the other major leap. Previous iterations produced completely silent sandboxes, requiring post-hoc sound effects or external synthesis. Genie 4 produces native audio matched to the world, delivering environmental ambience alongside voices for interactive characters directly within the generated stream. It follows broader efforts across Mountain View to streamline automated creative tools, much like how the Google AI Studio managed agents hub unified multi-step autonomous developer workflows.

Google Genie 4: Key specifications and features

Under the hood: Gemini 3.8 Flash and the two-minute limit

Underpinning the physics and behavioral logic of Genie 4 is Gemini 3.8 Flash, which coordinates scene state updates and interactive pacing. Google has also expanded simulation memory, allowing continuous experiences to run for about two minutes—double the one-minute limit of Genie 3.

Longer runs come with familiar generative weaknesses. Google explicitly warned that object permanence can degrade over extended runtimes, meaning items in the background or previously visited locations may morph or disappear as the simulation progresses. For now, the company targets Genie 4 as a rapid prototyping sandbox for game mechanics, robotic training scenarios, and film pre-visualization rather than a full replacement for traditional rendering engines. As consumer interest shifts toward specialized utility, as seen when Google dropped Gemini 4 Argon, Genie 4 serves as a technical showcase for generative physics and spatial reasoning.

Availability and pricing

Genie 4 is currently restricted to a closed research preview for Google's trusted testers. Google has opened a waitlist to test the system inside Project Genie.

A wider commercial release is scheduled to arrive 'coming soon' for Google AI Ultra subscribers. Google has not yet specified whether access will incur separate usage fees or remain bundled into existing subscription tiers.

While persistent memory across extended play sessions remains an ongoing challenge for generative models, Genie 4 represents a notable step toward controllable, synthetic interactive media.