[Video] Google DeepMind ‘Genie 4’ unveiled… Real-time creation of virtual worlds and sounds

(Seoul = Yonhap News) An artificial intelligence (AI) model that creates a virtual world in real time according to the user's movements and generates music and conversation at the same time has been unveiled.

Google DeepMind released a new world model, 'Genie 4', for research purposes through its official site on the 27th (local time).

The world model is an AI that learns real-world movements and physical laws to predict what will happen next based on the user's actions and creates a virtual environment.

The released video shows users moving around the virtual space using a keyboard and mouse, changing the weather and camera position, and adding new people and objects using text commands.

Rather than unilaterally playing back the completed scene like existing AI videos, it is characterized by changing the surrounding environment in real time by reflecting the user's operations and commands.

Genie 4 is designed to express trees swaying in the wind, water flow, object collisions and aerodynamics, and the virtual character is also designed to move and react according to surrounding situations.

Along with the video, music, ambient noise, and sound effects are also created in real time. The virtual character can also communicate based on the user's actions or events occurring around them.

However, the time a user can continuously be active in the virtual world created by Genie 4 is still only about a few minutes. As time passes or complex commands are repeated, consistency may decrease, such as the location and appearance of objects changing and letters becoming broken. Spatial and sound accuracy also needs improvement.