Google announces Genie 4 world model with longer AI-generated interactive experiences

What Google announced

Google has taken the wraps off Genie 4, the latest version of its AI "world model" for generating interactive experiences on demand. According to The Verge, the company announced the model on October 27 and is framing it as a clear step up from Genie 3, with longer outputs, audio support, and the ability to steer a simulation while it is running.

Four headline changes

As reported by The Verge, Google says Genie 4 arrives with four main improvements:

• It produces more realistic worlds than its predecessor.
• Users can modify an experience in real time with a natural language prompt, rather than restarting from scratch.
• The model can generate audio, not just visuals.
• It uses Gemini 3.8 Flash under the hood, which Google says helps drive the simulations.

Output length has also doubled: Genie 4 can create experiences of roughly two minutes, up from about one minute for Genie 3.

What it is for, and where it stumbles

Google is positioning the model as a practical tool rather than a pure research demo. The Verge reports that the company sees applications in pre-visualization for filmmakers, prototyping game ideas, and building training scenarios for robots.

There is an acknowledged limitation, though. Google warns that as simulations run longer, details can drift and object permanence may break down — meaning the system may lose track of what exists in a scene and where it is. That is a familiar weak spot for generative world models, which have to maintain a coherent environment without a conventional game engine to enforce the rules.

Availability

For now, access is narrow. Genie 4 is launching first as a "research preview" for "trusted testers," which means The Verge's reporter has not been able to try it and cannot yet say how it compares in practice to the Genie 3-powered Project Genie tool, which he previously used to generate crude Nintendo-style knockoffs.

Broader access is planned. Google says Genie 4 will be "coming soon" to Google AI Ultra subscribers, and it has opened a waitlist for people who want to try the model inside Project Genie.

Why it matters

World models are one of the more consequential bets in AI research right now. Instead of predicting text or static images, they learn the dynamics of environments — physics, cause and effect, how a scene evolves over time — and can generate playable, interactive spaces directly from a prompt. If the approach scales, it could sharply reduce the cost of prototyping games, film pre-visualization, and robot training, all areas where building even a rough interactive environment today takes real time and money.

The doubled clip length and the addition of real-time, prompt-based editing suggest Google is pushing toward usable, session-length experiences rather than short tech demos. But the company's own warning about drifting details points to the core unsolved problem: consistency. Until object permanence holds up over longer runs, these generated worlds are more likely to serve as fast prototyping and pre-visualization aids than as finished products — which is, notably, exactly the framing Google has chosen for Genie 4.