Audio, physics and people in Genie 4
Genie 4 is DeepMind's first world model with native audio. It generates it in real time, matched to the simulated environment: music, background sounds and interactive dialogue. Audio elements can be indicated directly in the prompt. DeepMind promises that dialogue will adapt to changes in the simulation.
Physics was also supposed to improve. The model handles aerodynamics and collisions better, and characters look less artificial and react more naturally to their surroundings. During longer sessions, details and objects begin to drift, and DeepMind admits it has not yet dealt with the credibility of audio or the legibility of text within the world.
Weather, camera angle, new people and objects can be changed in real time without stopping the scene. Coherence is maintained for up to two minutes, twice as long as in Genie 3.
For now, Genie 4 is available as a research preview for trusted testers.
Gemini 3.8 Flash as the simulation's director
Simulation control in Genie 4 is handled by Gemini 3.8 Flash, which in this configuration acts as the directing agent. The model breaks goals into steps and evaluates each generated frame, so the simulation does not fall off track when the user changes the scenery. TechCrunch describes it as the director: Gemini plans and checks, Genie creates the visuals.
In a research experiment linked to Gemini Robotics, the Gemini 4 Argon model broke tasks into steps that Genie 4 simulated in real time. For DeepMind, this is an argument that robot scenarios can be tested in simulation before they reach the real world. Genie 4 worlds remain coherent for only about two minutes. After that, the model forgets where it started.
For game, simulation and video creators
For commercial applications, DeepMind points to games, interactive teaching and robotics. Testers who were first given access to the model see yet another direction: film prototyping. Google DeepMind engineer Diego Rivas argues that real-time mode is most valuable where rapid ideation matters, for example at the storyboard stage.
Google also offers Veo, a model for generating video clips with synchronized audio, popular among filmmakers. Rivas notes that Genie 4 is still at an early stage and it is not yet known what the full product experience will look like.
DeepMind itself acknowledges the limitations. The longer the simulation runs, the more details and object persistence may change. The issues of audio accuracy and the legibility of text appearing in the generated world have also not been resolved.
Access will be gradual. The model will reach Google AI Ultra subscribers as part of Project Genie, as well as Gemini API and Google AI Studio partners. Prices have not yet been announced. The consumer version is to be introduced in stages in the most expensive AI Ultra plans, from 100 dollars per month.
Who Genie 4 concerns
Genie 4 will first reach trusted testers, and in the following weeks Google AI Ultra subscribers in Project Genie as well as Gemini API and Google AI Studio partners. Prices are not yet known; AI Ultra plans start at 100 dollars per month.
Sources
- TechCrunch: Google’s Genie 4 world model adds audio, more realistic worlds, and real-time steering
- Ars Technica: DeepMind unveils Genie 4 world model promising more realistic physics and humans
- The Verge - Tech: Google’s next Genie AI model generates more realistic experiences
Article prepared with the help of artificial intelligence based on: TechCrunch, Ars Technica, The Verge - Tech. Facts checked automatically against these sources. The Wiadomości PRO newsroom is responsible for this publication.




Comments (0)
Loading comments...