21:48Machine-translated from Azerbaijani by AIRead in Azerbaijani
Google DeepMind Unveils Genie 4 Model for Creating Virtual Worlds
On October 27, Google DeepMind released Genie 4, a world model that uses artificial intelligence to create interactive virtual environments, for research testing. A world model is a system that simulates the appearance, movements, and rules of a virtual space. The new version creates more realistic-looking scenes and responds to users’ written prompts during simulations. For now, the system is available only to selected testers.
Using written prompts, users can change the weather and camera angle or add a person or object to the scene. Genie 4 also generates music, ambient sounds, and interactive dialogue appropriate to that world. The model can keep a scene stable for up to 2 minutes. With Genie 3, this duration was 1 minute. Physical movements are more convincing, and the behavior of human characters also appears more natural. Gemini 3.8 Flash runs behind the simulation and helps construct the scene. Gemini acts as a director that plans the steps and reviews the generated frames, while Genie 4 creates the visuals. In a Gemini Robotics experiment, Gemini 4 Argon divided a task into stages, and the new world model simulated those stages in real time. This makes it possible to test robots’ movements in a virtual environment before conducting real-world trials. Google created the system to test game ideas and develop interactive education, robot training, and preliminary visuals for films. Filmmakers can quickly test different options while working on ideas and scene plans. One educational example involves virtually entering a historical painting and discussing the event with the characters depicted in it. In the future, the company plans to use Gemini Omni to turn such prototypes into finished creative material, but this capability is still a long way off. Veo is also a separate model that creates HD clips from text, images, and video, with audio synchronized to the visuals. The current version of the model still has some shortcomings. As a simulation continues, details may shift position, objects may disappear, and sounds may not fully match the scene. On-screen text can also sometimes become unreadable. Genie 4 will become available to Google AI Ultra subscribers through Project Genie in the coming weeks. Early-access partners for the Gemini API and Google AI Studio will also be able to use the system. Google is accepting applications for the waiting list.
Using written prompts, users can change the weather and camera angle or add a person or object to the scene. Genie 4 also generates music, ambient sounds, and interactive dialogue appropriate to that world. The model can keep a scene stable for up to 2 minutes. With Genie 3, this duration was 1 minute. Physical movements are more convincing, and the behavior of human characters also appears more natural. Gemini 3.8 Flash runs behind the simulation and helps construct the scene. Gemini acts as a director that plans the steps and reviews the generated frames, while Genie 4 creates the visuals. In a Gemini Robotics experiment, Gemini 4 Argon divided a task into stages, and the new world model simulated those stages in real time. This makes it possible to test robots’ movements in a virtual environment before conducting real-world trials. Google created the system to test game ideas and develop interactive education, robot training, and preliminary visuals for films. Filmmakers can quickly test different options while working on ideas and scene plans. One educational example involves virtually entering a historical painting and discussing the event with the characters depicted in it. In the future, the company plans to use Gemini Omni to turn such prototypes into finished creative material, but this capability is still a long way off. Veo is also a separate model that creates HD clips from text, images, and video, with audio synchronized to the visuals. The current version of the model still has some shortcomings. As a simulation continues, details may shift position, objects may disappear, and sounds may not fully match the scene. On-screen text can also sometimes become unreadable. Genie 4 will become available to Google AI Ultra subscribers through Project Genie in the coming weeks. Early-access partners for the Gemini API and Google AI Studio will also be able to use the system. Google is accepting applications for the waiting list.