Google DeepMind releases Genie 4: sound, realistic worlds, and real-time edits

What's new in Genie 4

On Tuesday, October 27, Google DeepMind released a new world model, Genie 4, in research preview status. According to TechCrunch, the model creates the most photorealistic and interactive worlds of any previous version: users can enter text prompts right during the simulation, changing the scene on the fly — for example, the weather, the camera angle, or adding new people or objects. For the first time, the model also generates sound in real time: music, background noise, and interactive dialogues matching the simulated world.

Progress over Genie 3

The previous version, Genie 3, entered research preview just over a year ago. Genie 4 maintains a consistent scene for up to two minutes — twice as long as its predecessor — and offers improved physics and more humanlike characters. The team acknowledges that the model still has shortcomings: details and objects can 'drift' during long simulations, and sound accuracy and text legibility within the world remain unsolved.

Gemini integration and possible applications

Genie 4 is more tightly integrated with the Gemini model, which acts as a 'director': it plans steps and checks frames while Genie creates the visual part. In an experiment related to Gemini Robotics, the Gemini 4 Argon model broke tasks into steps that Genie 4 simulated in real time — showing a possible way to test robot scenarios in simulation before validating them in the real world. 'We will increasingly see hybrid world models, where Gemini writes code and text, and Genie takes that in real time and simulates reality at different levels of abstraction,' said Google research director Tejas Kulkarni at a briefing.

Who it's aimed at

Among the stated commercial applications are games, interactive learning, and robotics; one speculative learning example is stepping into a painting with a historical scene and talking to its characters. Trusted testers who get early access to Google products point to a new use case: film prototyping. 'The real-time aspect is most valuable where it matters to preserve the flow state in the creative process,' noted Google DeepMind engineer Diego Rivas, citing ideation and storyboarding phases. Rivas added that over time users will be able to use Gemini Omni — a multimodal model that generates video from images, audio, and text — to turn prototypes into finished creative materials, although, according to him, such capabilities remain a distant prospect. Google also has Veo — a video generation model with synchronized sound that is gaining popularity among filmmakers.

Context: consolidation of the world model market

The release of Genie 4 comes amid intensifying competition and consolidation in the industry. A couple of months ago, video startup Runway released its second world model, GWM Worlds 2 — a playable model that dynamically responds to user actions and also generates sound. At the same time, the market is being bought up by major players: Fei-Fei Li's World Labs is buying AMD, Amazon is in talks to acquire Decart for roughly $7 billion, and Rivian acquired Third Dimensionto gain in-house simulation capabilities for testing autonomous driving. Genie 4 is another step by Google to hold its position in this increasingly crowded space.

How to get access

Genie 4 is currently available as a research preview for trusted testers. In the coming weeks, the model will become available to Google AI Ultra subscribers in Project Genie — Google's tool for generating interactive worlds from text or image prompts — as well as to early-access partners of the Gemini API and Google AI Studio. At the briefing, Rivas emphasized that the product is still at an early stage and it is unclear what the final user experience will look like.