Google launches Genie 4, a world model with audio and real-time scene editing
Reported by The Verge, Ars Technica, TechCrunch and 14 more outlets17 sources ↓
Google introduced Genie 4, which generates interactive worlds with audio and lets users change scenes through prompts while simulations run. It is available to select testers, with broader access planned for Google AI Ultra subscribers and some developer partners.
Key points
- Genie 4 can generate interactive worlds with audio and respond to prompts that change a scene in real time
- Google says scenes can remain consistent for up to two minutes, twice Genie 3’s limit
- The research preview is open to select testers; a wider rollout is planned for Google AI Ultra subscribers and developer partners
- Google says details and objects can drift in longer simulations, and accurate audio and readable in-world text remain challenges
Sources
- The VergeRead ↗Google’s next Genie AI model generates more realistic experiences
- Ars TechnicaRead ↗DeepMind unveils Genie 4 world model promising more realistic physics and humans
- TechCrunchRead ↗Google’s Genie 4 world model adds audio, more realistic worlds, and real-time steering
- The DecoderRead ↗Google Deepmind's Genie 4 puts a modern spin on agent designs that predate LLMs
- NewsCordRead ↗Google DeepMind Unveils Genie 4 World Model With Native Audio And Real-Time Steering: 10 outlets compared
Details
Genie 4 generates scenes frame by frame rather than producing a finished video, responding to keyboard controls and text prompts as users explore. TechCrunch reports that users can change elements such as weather or camera angle and add people or objects during a simulation. The model generates ambient sound and music, as well as dialogue that can respond to events or user input, according to Google DeepMind.
Google says Genie 4 can keep a scene consistent for up to two minutes, compared with one minute for Genie 3. Gemini models help plan events and check whether steps in a simulation worked. In a robotics research demonstration described by The Decoder and TechCrunch, Gemini 4 Argon broke a task into steps and Genie 4 simulated Apptronik’s Apollo humanoid robot playing checkers; no physical robot was involved.
The model is available as a research preview to select or trusted testers. Google plans to make it available soon through Project Genie to Google AI Ultra subscribers and to early-access partners using the Gemini API and Google AI Studio. Google says generated video and audio will carry SynthID watermarks.
Google DeepMind describes Genie 4 as an early experimental model. The Decoder also says Google is still researching complex physical interactions, accurate real-time sound and legible text; TechRadar argues that visual realism alone does not demonstrate general intelligence.
Written from coverage as of
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh