Google Unveils Genie 4, a Real-Time World Model With Audio

Google has announced Genie 4, the latest step in its push into world models. The system generates interactive environments, essentially digital worlds, from written prompts, and it does so in real time. That immediacy is what gives the model its flexibility, and the newest member of the Genie family brings upgrades on both the visual and the audio side.

Genie 4 is available from today as a research preview to a limited group of users. It is reportedly set to reach Google AI Ultra subscribers through Project Genie soon. For now, then, the model is in the hands of a small audience, with a wider rollout to subscribers expected in the near future.

Real-time environment generation and memory

The defining trait of the Genie series is its ability to turn a user's text prompt into an interactive environment. Rather than delivering a finished image or clip, it builds a world that can be explored and that responds to commands, which sets this class of system apart from conventional generative AI tools. Genie 4 keeps that approach, building its environments in real time, and generating them on the fly is the foundation of the flexibility the system offers.

The new version is also said to have a memory. This lets the model account for changes made in the environment when it generates subsequent frames. Users can move around the world while giving commands that the model carries out, and characters or vehicles of their choosing can be added in a controllable way. The generated world thus becomes a space shaped by the user's input, with changes carrying over into what comes next.

Audio generation and more realistic physics

One of the most notable additions in the new version is audio. Genie 4 can generate music, ambient sound and speech in real time, so the worlds it creates offer a richer experience that is no longer purely visual. Producing sound in real time, just like the imagery, stands out as a significant step beyond earlier versions.

The model also produces results that are closer to reality in its physics calculations. The new version is said to allow more interactive and realistic environments, raising the bar a notch. More faithful physics lets objects and characters in the environment behave in a more believable way.

Potential uses and availability

According to the information available, Genie 4 has significant potential in a wide range of areas, from game and film production experiments to specialized training simulations. The ability to produce environments instantly and to add controllable elements to them underpins that versatility. The model is therefore seen as more than a technology demo, a tool that could be tried out across different sectors.

The system remains at the research preview stage and is open only to select users. Reaching a broader audience depends on its release to Google AI Ultra subscribers through Project Genie. No firm date has been given for that step; the sources say only that the process will begin soon.