Google began rolling out Genie 4 on Nov. 18 to eligible subscribers of Google AI Ultra, its most expensive AI plan. They use it in Project Genie, an early test version of a browser tool that Google first opened earlier this year. Genie 4 is a world model, an AI that learns how the world behaves so it can predict what happens next. Here, that means Genie 4 draws each moment of a scene as the user moves through it, rather than playing back a finished video.
Google says users can change the world on the spot by typing a new prompt as they explore it: shift the weather, drop in a character or change what is happening. Genie 4 also makes its own sound as the scene plays out, from background noise to characters who speak, which Google says is a first for its world models. Google first showed Genie 4 on Oct. 27, when only invited testers could use it.
The update also adds a helper that builds a world from a conversation, and a gallery where people can publish their worlds for others to explore and remix. Google says a published world shows in the gallery with the name on the account that made it. When users are signed in, Google also collects what they do in Project Genie and the worlds the tool makes for them. The older Genie 3 stays in the tool for longer runs, now up to three minutes at a time instead of 60 seconds.
For now, Project Genie is open to Ultra subscribers, and subscribers to Google AI Pro, the $19.99-a-month plan in the U.S., get in only from a waitlist that Google says it will open gradually. Google's plans page lists two Ultra plans, at $99.99 and $199.99 a month in the U.S., and says Project Genie comes only with an Ultra plan, in more than 140 countries. Users must be 18 or older, and daily limits apply that may vary by plan and demand. Google says there is a delay between pressing a control and the world reacting, and users may have to wait their turn when demand is high. For now, making worlds does not use up the AI credits that come with the plans.
Google says all video and audio made in Project Genie carries SynthID, its AI watermark. That is a hidden signal an AI tool weaves into an image, video or audio clip as it makes it, which software can spot but people cannot see or hear. Google also says it screens what people type and what the tool makes for safety.
Google calls Genie 4 an early research model. It says details and objects in a world can shift or drift over time, especially in longer sessions, after several complex edits or as users move around. Google says it is still working on complex physical interactions, accurate sound, readable text and keeping a world consistent for a long time. Google's announcement does not say what Genie 4 will cost after this test. In October, Google said Genie 4 would also come to some developers in an early test of its Gemini API.
Update (Nov. 19): A day after this story, Google made Genie 4 available to developers through the Gemini API, its paid service for building Google's AI into their own apps, and through Google AI Studio. Google's pricing page says it comes only with paid API use and works out to about 10 cents a second, or about $6 a minute, while a world is running.