Google DeepMind just dropped Genie 4, an AI world model that generates photorealistic, interactive environments with real-time audio and mid-simulation steering—and the same corporation that decides which search results you see and which voices get erased from its platforms now holds the machinery to simulate reality itself.
Genie 4 lets users type prompts to reshape a scene on the fly—change the weather, move the camera, add people or objects—while the model generates every frame in real time. It produces ambient sound, music, and interactive dialogue that adapts as the simulation changes. Google's Gemini model acts as the "director," planning steps and evaluating frames to keep the simulation coherent, according to TechCrunch. Scenes hold consistency for about two minutes before the model loses the thread—double Genie 3's limit, per both TechCrunch and Ars Technica.
The physics are better too. DeepMind says Genie 4 handles aerodynamics and collisions more realistically, and human characters move more naturally. Ars Technica noted that Genie 4 is the first DeepMind world model with native audio, including what DeepMind promises will be interactive speech that adapts as you alter the simulation.
TechCrunch framed the release as Google staying competitive in a "crowded field," noting that Runway released its GWM Worlds 2 model months ago and that the industry is consolidating fast—AMD is acquiring Fei-Fei Li's World Labs, Amazon is in talks to buy Decart, and Rivian just bought Third Dimension. Ars Technica, meanwhile, led with how Genie 3 "sent gaming stocks tumbling" and emphasized that DeepMind sees world models as a stepping stone to artificial general intelligence, not just a video game tool.
What neither outlet dwelled on is the power concentration. Google already dominates search, owns YouTube, and has a documented history of suppressing content it deems undesirable—from throttling conservative outlets to deplatforming voices that challenge institutional narratives. Now that same entity is building tools that generate simulated worlds and could train physical robots. DeepMind's own research experiment tied Genie 4 to Gemini Robotics, simulating robot scenarios before real-world execution, TechCrunch reported.
The commercial pitch is gaming, film prototyping, and "interactive learning"—one example from Google involves entering a painting of a historical scene and chatting with its subjects. But who writes the history those simulated subjects recite?
Access remains gated. Google is rolling Genie 4 out to "trusted testers" first, then to developers via the Gemini API and AI Studio. Ars Technica was blunt: "It won't be cheap." Pricing hasn't been announced. Google DeepMind engineer Diego Rivas stressed it's still "early days" and that the end-to-end product experience isn't clear yet, according to TechCrunch.
A corporation that filters what billions of people can find, say, and see online now builds the tools to generate the very environments people will inhabit virtually—and eventually, the simulations that train the machines that move through the physical world. The question isn't whether Genie 4 is impressive. It is. The question is who gets to steer it, and who gets locked out.
