Google Launches Genie 4 - The AI That Builds And Runs Entire Worlds In Real Time

 

Google has launched Genie 4, a powerful new Artificial Intelligence (AI) model that can build and run entire virtual worlds in real time. It does not just create images or videos; it simulates living, breathing environments where things move, sounds play, and characters speak, all at the same time.

Genie 4 marks a significant step in DeepMind's AI research. For the unversed, DeepMind is Google's AI wing that handles the development and research of this technology.

Google describes Genie 4 as a world model. This means that it does not simply show the user a scene — it runs it, moment by moment, responding to whatever they do next.

 

Currently, Genie 4 is in research preview for trusted testers. Google is gathering feedback before a wider release. The first consumer access will come through Google AI Ultra plans via Project Genie, with early developer access through the Gemini API and Google AI Studio.

A public waitlist is now open.

 

Genie 4 is built on Google's Gemini family of AI models. Gemini Omni handles how the world looks and behaves, grounding the simulation in real-world appearance and physics. Other Gemini models handle reasoning, keeping track of what is happening in the scene and making sure events unfold in a logical, believable way.

Together, they allow Genie 4 to do something genuinely new, which is to simulate what happens next.

Users can set a scene in motion, and Genie 4 works out how the world reacts, how it sounds, and how the people within it move and speak.

 

Users can control Genie 4 worlds using standard keyboard inputs and simple text prompts in real-time. If a user wants to change the weather, change the camera movement mid-scene, bring a new character into an environment, and more, they simply have to type it. Gemini reads users' instructions, adapts them to fit the current scene, and applies them in real time.

One of the biggest advances in Genie 4 is audio. This is the first version of the model to generate sound natively, meaning every simulated world produces its own ambient noise — wind, waves, crowd sounds — automatically, without any added effects.

 

Genie 4 can generate real-time interactive dialogue. Characters in the simulation speak, react, and respond to what is happening around them. This is not pre-recorded or scripted. It adapts on the fly, in the moment.

Google Launches Genie 4 - The AI That Builds And Runs Entire Worlds In Real Time

Genie 4 delivers enhanced photorealism, fluid movement, and native audio. All accompanying sounds featured above are produced directly by Genie 4. (Image Credit: Google DeepMind)

 

Genie 4 also works alongside Gemini 3.8 Flash as a guiding agent. As the world runs, Gemini watches the scene, breaks tasks down into steps, and checks whether each step has been completed correctly. If the simulation changes or a user steps in, Gemini adjusts its plan immediately.

This makes it possible to simulate complex, multi-step events — things that follow a logical sequence of cause and effect — with much greater reliability than before.