On this page · 4 sections ▾
Google DeepMind released Genie 4, its latest world model for generating interactive simulated worlds, as a research preview on October 27. The model adds native audio, lets users change a scene with text prompts while it runs, and keeps scenes consistent for up to two minutes. Access is currently limited to trusted testers.
Key facts
- Release: research preview for trusted testers, announced October 27, 2026.
- Scene length: up to two minutes, double the limit of Genie 3, according to TechCrunch.
- Audio: music, ambient noise and dialogue that reacts to the user, generated natively.
- Next access: Google AI Ultra subscribers in Project Genie, plus early partners in the Gemini API and Google AI Studio, in the coming weeks. Ultra plans start at $100 a month, per The Next Web, which cites Ars Technica.
- Waitlist: open in Project Genie, with no date given.
What changed from Genie 3
Users can edit a running simulation in natural language, such as changing the weather or camera angle, or adding a person or object. Android Authority reports it is Google's first real-time world model with native audio, and that Google says it simulates aerodynamics and collisions and produces more natural human movement.
Gemini is tied in more closely. Android Authority says Gemini 3.8 Flash acts as a guiding agent that breaks a goal into steps and watches the scene as it is generated, adjusting if the user intervenes. The Verge also names Gemini 3.8 Flash as part of the simulation setup. Gemini Omni, per Android Authority and The Next Web, anchors how worlds look and behave.
In a robotics experiment, Gemini 4 Argon planned a task that Genie 4 simulated: Apptronik's Apollo humanoid playing checkers, according to The Next Web. TechCrunch describes the same experiment as a possible way to test robot scenarios in simulation before real-world trials.
Why it matters
Google pitches the model for gaming, interactive learning and robotics. Testers are also pointing to film prototyping. Google DeepMind engineer Diego Rivas told reporters the real-time aspect matters most for ideation and storyboarding, TechCrunch reports. The Next Web adds that testers in film, journalism, education, gaming and engineering have tried previsualization and game prototypes.
The release lands as the world model field consolidates. TechCrunch notes that Runway released GWM Worlds 2 a couple of months ago, that AMD is acquiring World Labs, that Amazon is in talks to buy Decart, and that Rivian acquired Third Dimension for self-driving testing.
Limits and what to watch
Google acknowledges gaps. Details and object permanence can drift as simulations run longer or after several edits. Audio accuracy and legible in-world text are still in development. All video and audio carries SynthID, Google's invisible watermark, per The Next Web and Android Authority.
Rivas also stressed that the end-to-end product experience is not yet clear, TechCrunch says. Watch for the Project Genie rollout to Ultra subscribers, the Gemini API and AI Studio access, and whether Google connects Genie prototypes to tools such as Veo or Gemini Omni for final assets. The Verge notes it has not tried Genie 4, so independent hands-on results are still pending.
