Google Genie 4 adds real-time edits and audio to AI worlds

Google announced its Genie 4 world model with real-time natural language edits, audio generation, and experiences up to about two minutes. It is coming first to trusted testers, then to Google AI Ultra subscribers.

What happened

Google announced today its Genie 4 world model for generating interactive experiences using AI. The company says the model has four main updates: it allows for more realistic worlds, lets users update an experience in real time with a natural language prompt, can generate audio, and uses Gemini 3.8 Flash under the hood to help with its simulations. Genie 4 also can create experiences that are about two minutes long, up from one minute for the previous Genie 3 model.

The model is launching first as a “research preview” to “trusted testers.” Google has opened a waitlist where people can apply to try it in Project Genie, and the company says Genie 4 will be “coming soon” to Google AI Ultra subscribers. Google is pitching Genie 4 as a tool for pre-visualization for filmmakers, prototyping game ideas, and making training scenarios for robots. Google also warns that, as simulations run longer, details and object permanence can shift over time.

Key facts

• Google announced Genie 4 today as a world model for generating interactive AI experiences.
• The model has four main updates: more realistic worlds, real-time natural language edits, audio generation, and Gemini 3.8 Flash for simulations.
• Genie 4 can create experiences that are about two minutes long, up from one minute for Genie 3.
• It is launching first as a “research preview” to “trusted testers” and will be “coming soon” to Google AI Ultra subscribers, with a waitlist open.
• Google warns that longer simulations can cause details and object permanence to shift over time.

Our analysis

For creators and social teams, Genie 4 moves AI generation from a static output to an interactive media tool that can be edited while it runs. The change from one minute with Genie 3 to about two minutes with Genie 4 gives teams more usable room for prototype scenes, concept clips, and short interactive demos. Real-time natural language rewriting is likely the most operator-relevant change: instead of regenerating an entire environment to fix a prop or camera path, a user can prompt the model mid-experience. That suggests a workflow closer to directing than to rendering.

The audio addition also changes the production value of an AI-generated world. Previously, a silent generated scene needed separate voice, sound design, or music work. With audio built into the same generation step, creators can test a more complete mood or narrative before committing to a full production. Google’s pitch to filmmakers, game prototypers, and robotics trainers suggests the company is targeting pre-production and simulation use cases, not final consumer entertainment. The warning about object permanence and shifting details over longer runs is important: the model is still best treated as a fast sketch tool, not a reliable world engine for long-form content.

What it means for operators

• Join the Project Genie waitlist now if you run concept work for campaigns, games, or video; early access through “trusted testers” may shape the tool’s limits and best practices before wider rollout.
• Build a prompt library for real-time corrections. Because Genie 4 accepts natural language updates during an experience, test short, specific commands for object swaps, lighting changes, and scene direction.
• Use the about two-minute length as a prototype window for pre-visualization, game ideas, or training scenarios. Plan cuts before object permanence drift becomes visible in longer simulations.
• Treat audio generation as part of the concept phase, not final sound. Generate a mood or placeholder track to test narrative timing, then replace it with owned audio if needed.