Google DeepMind unveils Genie 4 world model
Google DeepMind released Genie 4 on Tuesday, TechCrunch and The Verge report. The world model builds photorealistic interactive scenes in real time with audio and mid-stream text steering. It opens as a research preview for trusted testers, with Google AI Ultra access later.
Google DeepMind released Genie 4, its newest world model, as a research preview on Tuesday, TechCrunch and The Verge report. The model builds photorealistic interactive environments in real time and adds native audio such as music, ambient sound, and dialogue. Users can steer a scene mid-stream with text prompts, for example changing weather, camera angle, or adding people and objects.
Genie 4 can hold a consistent scene for about two minutes, double Genie 3's limit, TechCrunch and The Verge reported. Gemini helps plan steps and check frames while Genie renders the world. The Verge said Genie 4 uses Gemini 3.8 Flash under the hood. Pitched uses include game prototyping, interactive learning, robot training, and early film pre-visualization. Google engineers said details and objects can drift as a run gets longer, and on-screen text and audio accuracy still need work.
Access starts with trusted testers. Google AI Ultra subscribers are next through Project Genie, and early Gemini API partners follow later, TechCrunch reported. The Verge said Ultra access is coming soon and that Google opened a waitlist. The launch lands as other world-model firms consolidate, TechCrunch noted, with rivals shipping playable audio-capable models of their own.
DeepMind is still early on the end-to-end product shape. For now the demo case is rapid ideation: stay in a creative flow, try a prompt, and see the world change without restarting a full render. That is a different promise from pure video generators that spit out short clips and stop.
What we know: DeepMind opened Genie 4 as a research preview with real-time audio, mid-stream text steering, and roughly two-minute scenes for trusted testers.
What we don't: When Ultra and API users will get full access, or how well object permanence and audio hold up outside Google's demos.
Sources:
- TechCrunch — https://techcrunch.com/2026/10/27/googles-genie-4-world-model-adds-audio-more-realistic-worlds-and-real-time-steering/
- The Verge — https://www.theverge.com/tech/1019076/google-genie-4-ai-model
Where outlets agree TechCrunch and The Verge report that DeepMind launched Genie 4 with more realistic interactive worlds, real-time audio, text steering, and a research-preview rollout.
Where they differ The Verge names Gemini 3.8 Flash under the hood, while TechCrunch describes Gemini as the director planning steps without that version label.