Google's Genie 4 reaches paying subscribers, unfinished edges and all

4 Sources

Share

Google has opened Genie 4 to eligible AI Ultra subscribers via Project Genie, the research project's first public availability after an October reveal limited to trusted testers

2

3

1

. It is a step worth watching because a world model that until now only researchers could touch is suddenly something paying customers can explore from a browser.

Source: Google

Source: Google

What a world model actually is

Genie works nothing like a game engine. It predicts each scene as you move through it, generating the look and sound on the fly rather than rendering fixed video or a level with settled rules

3

. That is what makes it flexible, and it is also where the trouble comes from, because a predicted world can be wrong in ways a built one cannot.

The headline capability this time is editing. Type a new prompt mid-session and the world changes around you in real time: the weather shifts, a character appears, the action turns a different way, all without restarting

2

4

. Google bills it as the first Genie version with real-time native audio, generating ambient sound and character dialogue that shift as the scene unfolds

2

3

.

Who can get in, and what they find

The door is narrow. Genie 4 requires the $199.99-a-month Google AI Ultra tier in the US, and per MakeUseOf the project is limited to the Ultra 20x plan, excluding the cheaper Ultra 5x

4

. AI Pro members go on a waitlist instead, with no announced date for access

3

4

.

Google says the additions respond to what early testers asked for: more control over what happens inside a generated world, and easier ways to build and share one

2

. A new agent builds worlds from a natural-language conversation, pulling together Google's models for visuals and background scores. Gemini can also join a session to watch the screen, offer suggestions or trigger events inside the world

2

3

. Google frames the agent and the gallery as making world-building feel like a conversation and turning exploration into something people share and remix

2

. A community gallery launches alongside, with curated worlds including Living Memory, built from personal photos you can explore, plus collaborations with a creative technologist and filmmaker Anand Gandhi

2

3

.

Source: MakeUseOf

Source: MakeUseOf

Genie 3 stays available too, with sessions extended from 60 seconds to up to 3 minutes, 360-degree panorama seeding and the ability to chain worlds together. Google positions Genie 3 as the pick for longer runs or panorama-seeded spaces

2

3

.

The unfinished edges

Google itself lists the problems still unsolved: camera latency, physical interactions that break under complexity, and audio that drifts out of precision

2

3

. Objects move between views, scenes lose permanence over time, and generated speech or text can come out wrong

2

3

. Per Decrypt, the performance improvements Google claims for Genie 4 come from its own demonstrations rather than independent testing

3

. Every video and audio clip generated in Project Genie carries a SynthID watermark, with inputs and outputs passing through safety screenings

2

3

.

So a paying subscriber today gets a world that edits live and speaks aloud. The price limits who gets to try it, and the company itself calls the whole thing research. Google describes Genie 4 as its most realistic and interactive world model yet

2

3

. The gap between that description and the company's own list of unresolved problems is the honest measure of where the technology stands. Google gave no session length for Genie 4 itself, so the one spec that would say how long a world can hold together is the one it left out

3

.

© 2026 TheOutpost.AI All rights reserved