Key points
- Google DeepMind is rolling out Genie 4 to eligible Google AI Ultra subscribers through its Project Genie prototype.
- Users can type new prompts while exploring to alter weather, add characters or redirect events in real time.
- Genie 4 is the first Genie model to pair its visuals with native audio generated in real time, according to the company.
- Genie 3 sessions now run up to 3 minutes per generation, up from 60 seconds.
Google DeepMind says it has started rolling out Genie 4 to eligible Google AI Ultra subscribers. The company describes it as the most realistic and interactive of its world models. A world model, in this sense, is an AI system that generates an environment a person can step into and explore. Here, people access these worlds from a web browser through Project Genie, an experimental research prototype.
According to the announcement, Genie 4 lets people type a fresh prompt while they move through a world and change it as they go, shifting the weather, adding a character or steering the action. The company says it is the first Genie model to generate native audio in real time to match its visuals, from ambient soundscapes to character dialogue.
The older Genie 3 is still available, with some upgrades. Each generation can now run for up to 3 minutes, up from 60 seconds, and users can link several worlds together or start a new one from a 360-degree panorama image. The release also adds a Project Genie agent, which uses Google's multimodal models to build a world from a conversational description, and a community gallery where users can publish and remix one another's creations.
DeepMind says both models are still frontier research. It is working on problems including camera lag, complex physical interactions, how precisely the audio keeps up in real time, and long-term scene permanence. According to the company, all generated video and audio carries SynthID watermarking, and every model input and output goes through safety screening. Google AI Pro members can join a waitlist.
Why it matters
Google DeepMind says the update makes Project Genie's generated worlds more responsive and lets users edit them live and share them. It puts world models that the company still calls frontier research in the hands of paying subscribers.
What we don't know yet
All claims come from Google DeepMind's own announcement, and it gives no independent evaluation. The company says open challenges include camera latency, complex physical interactions, real-time audio precision and long-term scene permanence.
Terms in this story
- World model
- An AI system that generates an interactive environment that a user can move through and act within.
- SynthID
- Google's watermarking technology for marking content as AI-generated.
- Scene permanence
- A generated world's ability to keep places and objects consistent over time as a user explores.
Sources and verification
DexScience articles are produced with AI assistance and an automated fact-check against the linked sources. Spotted an error? Tell us and we will correct it openly.


