Google DeepMind Introduces Genie 4 With Real-Time Simulation, Native Audio Generation

Google DeepMind on Tuesday announced Genie 4 as its latest world model. The Mountain View-based tech giant says it can create realistic, interactive simulations that respond to user inputs in real time. The world model is claimed to be capable of generating environments with natural movement, realistic physics, and native audio, including ambient sounds, music, and interactive character dialogue. As per Google, Genie 4 is currently being tested with trusted partners.

Google DeepMind's Genie 4 Details Revealed

Released in research preview, Google DeepMind describes Genie 4 as a world model that combines real-time interactive simulation with improved realism. It builds upon the capabilities of the broader Gemini family, with Gemini Omni being used to create realistic simulations in real-world appearance and behaviour. Meanwhile, other Gemini models contribute reasoning to maintain coherence and continuity.

The world model allows users to modify a running simulation through natural-language prompts. They can adjust the weather, move the camera, and introduce a new person or object.

Among the notable additions is native audio generation. Google claims Genie 4 to be its first real-time world model that is capable of generating audio directly, from wind blowing through trees to waves crashing on a beach, alongside music and other ambient noises. This, as per the company, allows simulated environments to include sounds that match their settings.

Apart from this, it can generate interactive dialogue where characters are conversing and responding to user inputs or events as they unfold. Google says this eliminates the need to rely solely on pre-generated audio, as Genie can dynamically adapt the speech to changes made within the simulation.

According to the tech giant, Gemini 3.8 Flash serves as a guiding agent during simulation, continuously observing the generated scene, breaking down goals into smaller steps, checking incoming frames, and verifying whether each step has succeeded. Meanwhile, Google DeepMind is also experimenting with Gemini 4 Argon and Genie 4 alongside its Gemini Robotics research.

There are, however, some limitations. To begin with, it is an early-stage research project, which means details and objects may change or drift over longer runs, after multiple complex prompt edits, or while navigating through a scene. On the security front, Google claims Genie 4 has undergone internal safety evaluations and includes guardrails intended to prevent the generation of harmful or malicious content.

As per the company, all video and audio generated by Genie 4 will include its SynthID watermarking technology. It invisibly marks generated content with metadata that helps identify it as AI-created content for transparency and responsible use.

Genie 4 is being rolled out as a research preview to trusted testers. Google DeepMind will o expand access to Google AI Ultra subscribers through Project Genie, as well as early-access partners using the Gemini API and Google AI Studio.