Google DeepMind on October 27, 2026 announced Genie 4, the next generation of its world model, which generates explorable, interactive environments in real time from text or image prompts. The company is releasing it as a research preview to a limited group of trusted testers rather than as a public product, and it describes the model as an early, experimental step in its work on simulating the physical world.
The most visible change is sound. DeepMind says Genie 4 is its first real-time world model to generate audio natively, producing ambient noise, music and character dialogue alongside the visuals instead of layering sound on afterward. The company says characters can speak and respond to a user's inputs or to events unfolding in the scene, which it calls the hardest part of the audio work.
Genie 4 can also be redirected while it runs. Users navigate with standard keyboard controls and can type natural-language instructions mid-simulation to change the weather, move the camera, or introduce new people and objects. DeepMind says a Gemini model adapts those prompts to fit the surrounding scene, and that Gemini 3.8 Flash acts as a guiding agent in the loop, planning multi-step events, checking incoming frames and adjusting as the world changes. The company says Gemini Omni anchors the simulations in realistic appearance and behavior.
According to TechCrunch and The Verge, Genie 4 can hold a consistent scene for about two minutes, roughly double the limit of Genie 3, which Google released as a research preview a little over a year ago. DeepMind also claims more realistic physics, such as aerodynamics and collisions, and more natural-looking human movement. Those quality claims come from Google's own demonstrations; DeepMind has not published benchmark comparisons, and DigiEditorial has not been able to independently test the model.
Google is pitching the model at filmmakers who want to pre-visualize scenes, game developers prototyping ideas, photojournalists and educators exploring historical spaces from archival images, and teams building training scenarios for robots and AI agents. In one research experiment, DeepMind paired Genie 4 with its Gemini 4 Argon model, which broke a task into steps that Genie rendered in real time as a simulated humanoid robot played checkers.
DeepMind is candid about the model's limits. It says visual details and object permanence can shift as a simulation runs longer or after repeated edits, and that complex physical interactions, accurate real-time audio, legible text, spatial accuracy and long-term consistency remain open research problems. The company says Genie 4 passed internal safety evaluations before the preview and that all video and audio it generates carries Google's SynthID watermark.
Genie 4 is coming next to Google AI Ultra subscribers through Project Genie and to early-access partners of the Gemini API and Google AI Studio, with a waitlist open now. Google has not announced pricing or a firm date. DeepMind's October 27 blog post is the primary source for this report; DigiEditorial corroborated the announcement, availability and stated limitations against reporting from TechCrunch, The Verge and Ars Technica.