Frontier Flux Original
Google DeepMind unveils Genie 4, a world model with live text steering and native sound
Genie 4 builds explorable scenes in real time, adds ambient audio and talking characters, and lets users change a scene mid-session with text. It is limited to trusted testers for now, and DeepMind says details still drift in longer sessions.
Frontier Flux / AI-generated
Google DeepMind announced Genie 4 on 27 October 2026, a world model that generates an explorable scene frame by frame as you move through it. In its announcement, DeepMind says the new model adds native audio, including ambient sound, music and spoken dialogue from characters, and lets users change a running scene with text, such as switching the weather, moving the camera or adding a person.
Genie 4 is a research preview for trusted testers only. DeepMind says it is coming soon to Google AI Ultra subscribers in Project Genie and to early access partners in the Gemini API and Google AI Studio, with a waitlist open now.
Key facts
- Primary: Google DeepMind, 27 October 2026
- Independent: TechCrunch, The Verge and Ars Technica, 27 October 2026
- Gemini 3.8 Flash runs in the loop as a guiding agent, breaking goals into steps and checking incoming frames
- Session length: TechCrunch and The Verge report about two minutes of consistent scene, up from one minute for Genie 3
- Robotics test: Gemini 4 Argon split a checkers task into steps that Genie 4 rendered with Apptronik's Apollo humanoid
- All generated video and audio carries a SynthID watermark
- Pricing for the API has not been announced, and Ars Technica notes AI Ultra plans start at $100 a month
How to read it
The step up from Genie 3 is sound, people who move and talk more naturally, and live editing, not a finished product. DeepMind lists its own limits: details and objects can drift as a session runs longer or after several prompt edits, and audio accuracy, legible text and long-term consistency are still open problems. TechCrunch reports the team said the end product is not yet defined. For now the stronger case is pre-visualization for film and games and cheap training scenes for robots, which Google can test inside its own stack. Watch for API pricing and how long sessions last once wider access opens.
Independence
Frontier Flux is not affiliated with Google, DeepMind, TechCrunch, The Verge or Ars Technica. Capability claims come from DeepMind and have not been tested by Frontier Flux.