Google opens Genie 4 world model to AI Ultra subscribers
Google is rolling out Genie 4, real-time world editing and a community gallery inside Project Genie for eligible AI Ultra subscribers.
Google is opening its latest interactive world model to paying subscribers, turning Project Genie from an experimental playground into a more capable place to create, edit and share explorable digital spaces. The rollout is limited to eligible Google AI Ultra subscribers, while Google AI Pro members can join a waitlist. 1
What Genie 4 adds
Google says Genie 4 lets users change a world while they are exploring it. Text prompts can alter the weather, add characters or redirect the action, with native audio adapting to the scene through ambient sound and dialogue. The company describes Genie 4 as its most realistic and interactive world model yet, but those performance claims remain Google’s own rather than the result of independent testing. 1
The update also introduces a Project Genie agent that builds interactive experiences through conversation. Instead of stopping at a single text or image prompt, the agent can orchestrate the visual world, soundtrack and rules around a user’s description. Gemini models can watch the on-screen action and offer contextual suggestions or trigger events as a session unfolds. 1
Genie 3 gets longer-running worlds
Google is keeping Genie 3 in the model picker for users who want longer runs or panorama-based spaces. The updated model supports generations of up to three minutes, compared with 60 seconds previously, and can use 360-degree panorama images to ground a world. Users can also link worlds together and prompt real-time events, allowing a journey to move between different environments. 1
A gallery for sharing and remixing
Project Genie now includes a gallery of curated and community-built worlds. Google says users can publish creations for others to explore and remix, with examples ranging from a gothic manor mystery to an ant-colony experience. The company is also highlighting worlds built from personal photos, panorama-based scenes and small interactive experiments. 1
Google says generated video and audio carry SynthID watermarking, and that model inputs and outputs go through safety screening. It also acknowledges that camera latency, detailed physical interactions, real-time audio precision and long-term scene permanence remain active research challenges. That caveat matters: Genie 4 is an impressive demonstration of interactive generation, but Google is still describing it as frontier research rather than a settled platform. 1
Our opinion
Opening Genie 4 to subscribers is a meaningful shift because the interesting part is no longer simply generating a pretty scene. The pitch is a world that can respond when you change your mind, then become something other people can explore and remix. The limits are just as important as the spectacle: access is paywalled, the sessions are still short, and Google’s own examples cannot tell us how reliably the system handles complex prompts over time. Project Genie looks most useful as a creative toy and research preview for now, not a replacement for a game engine or a dependable production tool.