Google opens Genie 4 world model to AI Ultra subscribers
A revamped Project Genie adds real-time world editing with native audio, a creation agent and a community gallery, while Genie 3 gets three-minute sessions.

- What happened
- On 18 November Google began rolling out a new Project Genie experience, including Genie 4, to eligible Google AI Ultra subscribers.
- What’s new
- Genie 4 lets users change a generated world mid-session by typing prompts, and adds real-time native audio for the first time.
- Why it matters
- It is the first public access to Genie 4, Google's latest interactive world model, though Google still describes it as frontier research.
Google has started rolling out a reworked version of Project Genie, its experimental prototype for generating and exploring interactive worlds, to eligible Google AI Ultra subscribers. The 18 November update gives users their first access to Genie 4, alongside upgrades to Genie 3, a new Project Genie agent and a community gallery. Google AI Pro members can join a waitlist.
Google calls Genie 4 its most realistic and interactive world model so far. Users can type a new prompt while moving through a world to alter it on the fly, for example changing the weather or adding a character. According to Google, it is also the first Genie model to generate native audio in real time, ranging from ambient sound to character dialogue.
Genie 3 stays in the model picker for longer runs. A single generation can now last up to 3 minutes, up from 60 seconds, and worlds can be chained together. Genie 3 can also start worlds from 360-degree panorama images and accept prompts that trigger events during exploration.
The Project Genie agent builds a world's initial look and a matching soundtrack from a plain-language description, and Gemini can join a session to suggest directions or trigger events. Users can publish and remix worlds in the gallery.
Google says all generated video and audio carries SynthID watermarks and that inputs and outputs are safety-screened. It also acknowledges open problems, including camera latency, complex physical interactions, audio precision and keeping scenes consistent over time.