Key facts
- Google said Genie 4 lets users type instructions mid-session to change weather, add characters or steer events.
- Google said the model generates sound, from wind and footsteps to dialogue reacting to scene events.
- Google said Genie 3 stays available with sessions extended to three minutes and 360-degree panorama support.
- Google said it did not specify a session length for Genie 4 in the November 18 announcement.
- Google said AI Pro subscribers can join a waitlist; no date was given for their access.
- Google said Project Genie audio and video carry SynthID watermarks and inputs and outputs face safety screening.
Google said on November 18 that Genie 4, its latest world model, is rolling out through Project Genie, a browser-based experiment for generating interactive environments, to eligible AI Ultra subscribers. The company unveiled the model in October, when access was limited to trusted testers.
Rather than rendering a finished video from a prompt, Genie generates a scene as a user explores it. Google said that with the fourth model, users can type instructions midway through a session to change the weather, introduce a character or steer events without starting over. The model also creates sound alongside the visuals, from wind and footsteps to dialogue that reacts to events in the scene.
Google said the model improves how people move and how objects respond to physical forces, though the report noted those claims come from Google's own demonstrations rather than independent testing. Google acknowledges that objects can drift or change between views, complex physics remain difficult, and generated speech or text may be inaccurate. It said Project Genie's audio and video carry SynthID watermarks and that inputs and outputs undergo safety screening.
The update sits inside the Google AI Ultra subscription, which costs $199.99 a month in the United States. Google said eligible Ultra users can access the new experience, while AI Pro subscribers can join a waitlist. Google has not announced when Genie 4 will reach AI Pro subscribers.
Genie 3 remains available alongside the new model, with sessions extended to three minutes, support for 360-degree panorama images and the ability to connect multiple generated worlds into longer journeys. Early Project Genie sessions, which opened in January to adult Ultra subscribers in the United States, lasted 60 seconds. Google did not specify a comparable session length for Genie 4 in the November 18 announcement.
The report traces the line back to 2024, when the first Genie learned from more than 200,000 hours of gameplay footage and could turn a single image into a simple playable 2D environment. Genie 2 moved the research into three-dimensional spaces later that year, and Genie 3 took a further step toward continuous, real-time exploration.
