Genie 4 is now available
Google DeepMind has begun rolling out Genie 4, the newest iteration of its world model research project, to eligible Google AI Ultra subscribers. The update arrives under the Project Genie umbrella, an ongoing effort at DeepMind to build generative models capable of producing interactive, playable environments - not just static images or video clips.
World models like Genie are designed to learn the underlying structure of visual environments well enough to simulate them. Where earlier versions of Genie demonstrated the ability to generate 2D platformer-style worlds from a single image prompt, successive iterations have pushed toward richer environments, more consistent physics, and finer user control over how those environments behave. Genie 4 continues that progression, though the full scope of its technical improvements over its predecessor has not yet been detailed publicly.
The rollout is currently gated behind Google's AI Ultra subscription tier, the company's highest-tier AI access plan. This approach follows a pattern Google has used for other early-access AI features, using its subscriber base to manage capacity and gather real-world usage data before any broader release. It also signals that Genie 4 is being positioned as a premium, consumer-facing experience rather than a purely research-facing artifact.
The practical applications for a capable world model span several areas - game development prototyping, interactive storytelling, simulation for robotics training, and generative media more broadly. Whether Genie 4 brings meaningful advances in any of these directions will become clearer as more users gain access and share results. For now, the release confirms that DeepMind continues to treat interactive world generation as one of its core long-term research bets.


