Google DeepMind announced Genie 4 on October 27, and it's the clearest step yet from AI video toward AI places. You don't just watch a clip. You move through a scene with standard keyboard controls, type a change mid-simulation, and watch the world respond. Change the weather, move the camera, or drop a new character into the frame, and Gemini adapts your prompt to fit the scene around it.

The sound is what makes this release different from Genie 3. Genie 4 is DeepMind's first real-time world model to generate audio natively, so every world carries its own ambient sound, from wind in the trees to waves at a beach. Characters can also hold interactive dialogue that reacts to what you type or to events in the world, in real time.

How it stays on track

Under the hood, two systems work together. Gemini Omni anchors the simulation in real-world appearance and behavior, while other Gemini models supply the reasoning. Specifically, Gemini 3.8 Flash acts as a guiding agent in the loop: it watches the scene, breaks a goal into steps, checks incoming frames, and adjusts the plan when you intervene.

That pairing is also how DeepMind is testing robotics ideas. In one official example, Gemini 4 Argon breaks down a task into steps while Genie 4 renders Apptronik's Apollo humanoid playing checkers. The point isn't the game. It's that a robot could practice a multi-step job in a simulated place before anyone lets it near the real thing.

The limits, in DeepMind's own words

We checked the official post, and DeepMind is plain about the rough edges. Visual details and object permanence can still drift, especially in longer sessions, after several complex edits, or as you move around the scene. Complex physics, real-time audio accuracy, legible in-world text, spatial accuracy, and long-term consistency are all listed as ongoing research. TechCrunch reports scenes hold together for up to about two minutes, roughly double Genie 3, but that's press reporting, not an official spec, so treat it as a rough guide.

Our take

Here's the catch: you probably can't try it yet. Genie 4 is rolling out as a research preview for trusted testers, and DeepMind says it will come soon to Google AI Ultra subscribers in Project Genie and to early access partners of the Gemini API and Google AI Studio. There's a waitlist for Project Genie, but no general release date and no API pricing in the announcement, so don't take any price you see elsewhere as confirmed. All video and audio it generates carries SynthID, DeepMind's watermark, which is a sensible guardrail for worlds this convincing. For filmmakers sketching a scene, game teams prototyping an idea in seconds, or robotics researchers needing cheap practice scenes, this is genuinely exciting. For everyone else, it's a preview worth watching, not a tool to plan around yet.