Google DeepMind has launched Genie 4, the latest version of its experimental AI technology that builds fully interactive 3D environments on demand, with Google CEO Sundar Pichai calling it the company’s “most realistic and interactive world model.” Announcing the release on X, Pichai said the tool has begun rolling out as a research preview to trusted testers, and will soon reach Google AI Ultra subscribers through Project Genie, alongside early access partners via the Gemini API and AI Studio.
According to Google, Genie 4 constructs virtual spaces frame by frame as users explore them. Players can navigate with keyboard controls while issuing natural language commands to reshape the scene, an approach that sets the system apart from earlier models that required users to restart to make changes.
In a blog post, Google said the technology lets people “set an event in motion and steer scenes in real time using natural language, like moving the camera, changing the weather, or introducing new characters.” The company added that “every world also generates native audio on the fly, including ambient sounds, music, and interactive dialogue.”
The headline upgrade is audio. Genie 4 is DeepMind’s first real-time world model to generate sound alongside visuals, meaning every environment carries the ambient noises of its setting, from wind moving through trees to waves breaking on a beach. A connected Gemini model interprets typed instructions in context, and even monitors the world between user inputs to keep the scene progressing coherently. Users can now type commands mid-session rather than starting from scratch.
DeepMind describes human speech as the true frontier. Characters in Genie 4 converse and react to user inputs and in-world events in real time, while improvements to collisions, aerodynamics and body movement produce more lifelike human reactions and interactions. “The most dramatic leap is how Genie 4 simulates people, creating characters that move naturally, respond organically to their environments, and can be interactively steered in real-time,” the company wrote.
The model also holds a consistent visual scene for longer, lasting up to a few minutes, according to TechCrunch, with some reports citing two minutes. That is double the capacity of Genie 3, though continuity problems reportedly persist.
DeepMind framed the work as a step toward general artificial intelligence. “Understanding and simulating the world is a crucial aspect of building AGI,” the company noted, adding that Genie 4 expands its ability to create “safe, infinitely variable simulations” that could train robots and agents or unlock new forms of interactive entertainment.
Even so, Google cautions that the technology remains at an early research stage, despite demonstrations that showcase striking creative potential.
Original Source: https://www.ndtv.com/feature/google-deepmind-unveils-genie-4-all-about-real-time-world-building-with-built-in-sound-12250501#publisher=newsstand
Category:
Tags:
Publish Date: 2026-10-28 22:29:00
