Google Announces Genie 4, a World Model for Prompt-Generated Interactive Experiences

Google has announced Genie 4, a world model that generates interactive experiences from prompts, and will launch it first as a research preview for trusted testers. The announcement was reported on 2026-10-27 GMT. The Verge described the model as producing more realistic worlds than its predecessor.

A world model predicts how a simulated environment changes in response to user input, so the output can be explored rather than only watched. Users can change a running experience in real time with a natural-language prompt, rather than restarting generation. Genie 4 can also generate audio, and it uses Gemini 3.8 Flash to assist with its simulations. Experiences can run about two minutes, up from roughly one minute for Genie 3.

Google is pitching the model for three uses: pre-visualization for filmmakers, prototyping game ideas, and building training scenarios for robots. Each of these depends on a world that holds together over time, and Google has flagged that limitation directly. As Genie 4 simulations run longer, details and object permanence (the property that an object keeps its identity when it leaves view or the scene moves on) can shift. Objects may change or fail to persist as the scene continues.

Access is staged. Google opened a waitlist for trusted testers to apply to try Genie 4 through Project Genie, and says the model will come soon to Google AI Ultra subscribers.

Genie 4 builds on a line of work that Google DeepMind began publicly with Genie 2, which it described in December 2024 as a large-scale foundation world model. DeepMind's Genie 3 announcement in August 2025 described it as the lab's first world model to allow interaction in real time, with improved consistency and realism over Genie 2. In January 2026, Google introduced Project Genie as an experimental research prototype that lets users create, explore and remix their own interactive worlds. Later that year, Google said Genie is a general-purpose world model capable of generating diverse, interactive environments. Project Genie accepts text and images as input.

In May 2026, Google said it was connecting Project Genie with nearly 20 years of Street View imagery, so that generated worlds could be anchored to real places. That connection is the clearest sign so far of the direction the lineage has taken: away from purely synthetic scenes and toward environments grounded in mapped geography.

Markets reacted to the earlier release. Reuters reported that videogame stocks slid after Google's announcement of an AI model that turns prompts into playable worlds. Genie 4 is pitched for pre-visualization and prototyping rather than as a consumer game engine, but the overlap with interactive entertainment is the reason the January reaction drew attention.

In my view, the figure that matters most for the professional use cases is not the two-minute ceiling but the drift. A pre-visualization shot or a robot training rollout needs the same object to be the same object at the end of a sequence as at the start. Google's own caveat suggests that consistency over long horizons is still the open engineering problem. The real-time editing feature raises the stakes, because every mid-run prompt is another opportunity for the scene state to diverge.

The broader context here is that world models are being evaluated on two axes at once: visual and audio fidelity, and whether the simulated state stays coherent under interaction. Genie 4 addresses the first more visibly than the second. Its natural-language editing and audio generation extend what a single session can do. If preview testers find that the state holds across the target durations, the same architecture could give studios and robotics teams a lower-cost way to test scenarios before committing physical or production resources.