Google Genie 4 Brings Audio and Real Physics to AI Worlds — With a Catch

Google DeepMind has officially unveiled Genie 4, the latest evolution of its interactive world simulation model. Moving past simple visual generation, the new system introduces native audio support, improved physics for collisions and aerodynamics, and more organic human characters that react directly to user input.

How Gemini 3.8 Flash Keeps Simulations on Track

Under the hood, Genie 4 leverages Gemini 3.8 Flash as a guiding agent. According to Ars Technica, this underlying model breaks down user goals and evaluates each generated frame in real-time, preventing the simulation from completely breaking context when parameters change.

Despite this intelligence boost, Ars Technica points out a critical technical limitation noted by DeepMind: generated worlds still begin to lose their original coherence and forget their starting state after roughly two minutes of continuous exploration. This marks an improvement over the one-minute limit of Genie 3, but object permanence remains a challenge for longer sessions.

How It Works in Practice

Users can shape and alter the virtual environment on the fly using natural language prompts. If you want to change the weather, adjust camera angles, or spawn new objects, Gemini contextually adapts the input into the surrounding scene while generating ambient noises and interactive dialogue simultaneously.

Step 1: Apply for the waitlist through Project Genie or access it via early developer channels in Google AI Studio and the Gemini API.

Step 2: Input an initial text prompt or reference image to establish the starting frame of your virtual environment.

Step 3: Navigate using standard keyboard controls while issuing real-time text commands to alter weather, physics, or characters.

Is It Worth It for Your Workflow?

Google is pitching Genie 4 as a rapid prototyping tool for filmmakers, game developers, and roboticists looking to build complex training environments. However, access comes with a steep price tag.

The Verge confirms that the research preview is initially rolling out to trusted testers. Consumer access will be heavily gated behind Google's most expensive AI Ultra plans starting at $100 per month via Project Genie, with developer pricing through the Gemini API still unannounced.