Google DeepMind previews Genie 4 with audio and real-time world edits

Google DeepMind has introduced Genie 4, its latest world model, in a research preview for trusted testers, according to TechCrunch. The model turns prompts into interactive simulated worlds, now with real-time audio and the ability to steer a scene while it is running.

For creators, marketers and small studios, the important shift is not just better video-like output. Genie 4 points toward AI systems that can behave more like live environments than one-way clip generators, which could matter for storyboarding, learning content, game concepts and robotics simulation.

What Genie 4 adds to Google’s world-model push

Genie 4 is designed to generate more photorealistic and interactive worlds than Google’s previous version, Genie 3, according to the company. Users can type new prompts during a simulation to alter what is happening, including changing the weather, moving the camera or introducing new people and objects.

Google says the model can keep a scene consistent for up to two minutes, which TechCrunch notes is twice the duration of its predecessor. The company also says Genie 4 improves physics and character behavior, and adds audio that follows the simulated environment, including music, ambient sound and interactive dialogue.

That audio component is notable because many creative workflows still split visual generation, sound design and dialogue into separate tools. If the model can keep those elements aligned during an interactive session, it could reduce early-stage friction for teams exploring a setting or sequence. Google also acknowledged, however, that audio accuracy remains an unresolved issue.

Gemini takes a director role

Google is tying Genie 4 more closely to Google Gemini. In the setup described by TechCrunch, Gemini helps direct the process by planning steps and checking frames, while Genie handles the visual simulation.

In one research experiment connected to Gemini Robotics, a Gemini 4 Argon model split tasks into steps and Genie 4 simulated them in real time. Google presented this as a possible path for testing robot scenarios virtually before attempting them in the physical world.

Tejas Kulkarni, a research director at Google, described a future of hybrid systems that combine Gemini’s code and text generation with Genie’s ability to simulate reality at different levels of abstraction. For businesses, that suggests Google sees world models not only as media tools, but also as planning and testing layers for more complex AI systems.

Google already has Veo, its AI video-generation model for creating high-definition clips with synchronized audio from text, image or video prompts. Genie 4 appears to sit in a different lane: less about producing a finished linear clip immediately, and more about exploring an environment that can be changed in the moment.

Where creators may use it first

Google’s suggested commercial uses include gaming, interactive learning and robotics, according to TechCrunch. One example raised in the briefing was the idea of entering a painting of a historical scene and conversing with the figures depicted in it.

Trusted testers are also pointing to film prototyping as a possible use case. Diego Rivas, a Google DeepMind engineer, said during a press briefing that real-time interaction is especially useful when creators need to stay in a flow state, such as during ideation or storyboarding.

For an independent filmmaker, agency or educator, that makes Genie 4 most relevant at the rough-draft stage rather than the delivery stage. Possible early workflows include:

• testing a scene layout before commissioning final assets;
• exploring camera angles for a storyboard;
• trying variations of mood, weather or setting in a single session;
• building interactive learning demos around historical or fictional environments.

Rivas also said users could eventually use Gemini Omni, Google’s multimodal large language model, to turn prototypes into creative assets, but he stressed that those capabilities are still not near-term.

Runway and the limits to watch

Google is moving in a market that is already active. TechCrunch notes that Runway released GWM Worlds 2 a couple of months ago, describing it as a playable world model that reacts to user input and includes audio generation.

The broader world-model sector is also shifting. According to TechCrunch, World Labs is being acquired by AMD, Amazon has been in talks to buy Decart, and Rivian acquired Third Dimension to bring simulation technology in-house for self-driving testing. That activity shows why Google is pushing Genie forward, even while the product experience is not yet fully defined.

For readers tracking AI video, this follows other moves toward generated media with audio and longer-form output, including Grok’s AI-generated Odyssey preview and Kling AI 4.0’s audio and lip-sync update. Genie 4’s distinction is its focus on steerable, stateful worlds rather than only rendered clips.

Google also disclosed important caveats. Details and objects may drift during longer simulations. The company has not fully solved accurate audio, and text that appears inside generated worlds may not be reliably readable.

Availability is limited for now. Genie 4 is currently in research preview for trusted testers. In the coming weeks, Google plans to bring it to Google AI Ultra subscribers inside Project Genie, as well as early access partners using the Gemini API and Google AI Studio.