Story
Oct 27, 2026

Google’s Genie 4 Looks More Real Than Ready

DeepMind’s Genie 4 brings real-time steering, stronger physics and native audio to AI-generated worlds. But Google’s own rollout and technical caveats underline how far the technology remains from dependable simulation.

Google DeepMind unveiled Genie 4 on Tuesday as a research preview, pitching its latest “world model” as a system that can generate explorable environments frame by frame rather than relying on prebuilt game worlds. The company says users can alter weather, objects, camera angles and actions while a simulation is running.

Google CEO Sundar Pichai framed the release as its “most realistic and interactive world model,” highlighting real-time control over dynamic environments and native audio ranging from ambient sound to conversational characters.

The upgrade is substantive on paper. Genie 4 can hold a scene together for up to two minutes — twice Genie 3’s reported limit — while Gemini acts as a director, planning tasks and checking generated frames. DeepMind says better handling of collisions, aerodynamics and human movement could make the model more useful for robot training, interactive learning and rapid creative prototyping.

That ambition is already colliding with a crowded market. Runway recently released GWM Worlds 2, while consolidation is accelerating around world-model developers and simulation firms. In that race, Google’s pitch is not merely prettier AI video; it is a platform that could let Gemini map out a task and Genie simulate it before a robot encounters the real world. Research director Tejas Kulkarni described the destination as “hybrid world models,” combining Gemini’s code and text generation with Genie’s real-time simulation.

Yet the caveats are as important as the demo. DeepMind acknowledges that objects and details can drift during longer runs, while accurate audio and legible in-world text remain unresolved problems. Diego Rivas, a DeepMind engineer, stressed that it is “still early days” and that the end-to-end product experience is not settled.

For now, availability reflects that uncertainty. Genie 4 is limited to trusted testers, followed by Google AI Ultra subscribers and selected Gemini API and AI Studio partners. Consumer access will arrive in phases through Project Genie, with Google’s $100-a-month Ultra tier first in line and no firm timetable for wider release.