Google DeepMind’s Genie 4: AI Worlds You Can Explore, Hear, and Rewrite Live
Key takeaway: Google DeepMind announced Genie 4 on October 27, 2026, a world model that generates explorable, photorealistic worlds in real time — now with native audio, including ambient sound, music, and characters who talk back. Scenes stay consistent for about two minutes, and you can rewrite them mid-simulation with plain-language prompts.
Google DeepMind announced Genie 4 on October 27, 2026, as its new world model for real-time interactive simulation. It is in research preview for a small group of trusted testers, with wider access planned for Google AI Ultra subscribers in Project Genie and early access partners of the Gemini API and Google AI Studio. You can read the official DeepMind announcement here.
What Genie 4 actually is
A world model is not a video generator. A video model produces a clip you watch; a world model simulates a place you move through, predicting what happens next as you act. With Genie 4, you type a text prompt or provide a reference image, and it builds a world around you frame by frame, navigated with keyboard controls, that keeps going between your inputs.
DeepMind describes Genie 4 as natively multimodal. Music, ambient noise, and interactive dialogue are generated together with the visuals, in the same pass, rather than dubbed on afterwards. A lakeside scene carries the sound of water; a character in the scene can answer you in its own voice and react to events as they unfold.
What is new compared with Genie 3
Genie 3, launched in August 2025, could already generate navigable worlds but produced no sound and held a scene together for roughly one minute. Genie 4 roughly doubles that to about two minutes, with more photorealistic visuals, better physics, native audio, and free-form live editing. DeepMind calls the simulated people the most dramatic leap: characters now move more naturally, respond to their surroundings, and can be steered in real time.
| Feature | Genie 3 | Genie 4 |
|---|---|---|
| Audio | None | Native music, ambient sound, and interactive dialogue |
| Scene length | About 1 minute | About 2 minutes |
| Live editing | Limited | Change weather, camera, people, and objects mid-scene |
| Physics and people | Basic | More convincing aerodynamics, collisions, and natural human movement |
| Availability | Research preview | Research preview; Ultra and API partners next |
How live steering works
While a world runs, you can type a prompt to change the weather, move the camera, or drop a new person or object into the scene — without restarting. Gemini Omni anchors the simulation in how the real world looks and behaves, while Gemini 3.8 Flash acts as a guiding agent inside the loop: it observes rendered frames, breaks a goal into steps, and adapts its plan when you intervene.
What Genie 4 costs, and who can use it
Genie 4 costs nothing to read about today, because you cannot buy direct access yet. It is available only to trusted testers in research preview, and Google has not announced a standalone price or a firm public release date. The first wider routes are Google AI Ultra subscribers through Project Genie and early access partners of the Gemini API and Google AI Studio, with a waitlist open for Project Genie.
What you need before you can try it
The only prerequisite right now is a place in the queue. For testers, access runs through Project Genie or the Gemini developer tools, not a public app listing. For everyone else, the practical step is joining the waitlist and watching for the Ultra and API rollouts DeepMind says will come in the following weeks. No special hardware has been announced; the simulation runs on Google’s side.
Why it matters
World models are becoming the rehearsal stage for AI systems that will act in the real world. DeepMind is already testing Genie 4 with Gemini Robotics work, including simulated robot tasks, so agents and robots can practise in infinitely variable simulations before touching physical hardware. The same technology feeds game prototyping, interactive learning, and film previsualisation. It also changes AI training itself: if agents can learn inside cheap, generated worlds — like the dynamic AI agents that plan and adapt on the fly — developers need fewer expensive real-world trials, a theme also visible in emerging technologies in media.
Risks and limits to keep in mind
Two minutes is still short. Consistency can fray over longer runs, with small details drifting and object permanence slipping, according to early coverage — so Genie 4 is a research preview, not a production pipeline for games or films yet. Audio and in-world text accuracy also remain unsolved problems. On safety, generated worlds with talking characters make convincing synthetic scenes easier to produce, which is why Google says every Genie 4 clip will carry SynthID, its imperceptible watermark. That helps attribution, but it does not by itself stop misuse.
Knowns, unknowns, and what to watch
Known: Genie 4 is real, announced October 27, 2026, in research preview, with native audio, live prompt steering, roughly two-minute scenes, and Gemini-powered guidance. Unknown: the public release date, pricing beyond the existing Ultra and API routes, and how well quality holds up outside DeepMind’s curated demos. What to watch: the Ultra rollout in Project Genie, early developer reports from the Gemini API, and whether rival world models match the audio-plus-steering combination before Genie 4 goes public.
Analysis: a demo that points at a platform
Analysis — not a confirmed fact: Genie 4 reads less like a finished product and more like DeepMind staking a claim: whoever builds the best simulator may also build the best training ground for agents and robots. The two-minute limit is the honest constraint, and the race now is who stretches a coherent, audible, steerable world from minutes to hours. Until independent testers get their hands on it, treat the demos as a direction of travel, not a benchmark.
Sources
- Google DeepMind — Genie 4: A new generation of world simulation (October 27, 2026)
- SiliconANGLE — Google introduces cutting-edge Genie 4 world model (October 27, 2026)
- The Indian Express — What is Genie 4, Google’s new world model? (October 2026)
Frequently asked questions
What is Google DeepMind’s Genie 4?
Genie 4 is a world model announced on October 27, 2026, that generates explorable, photorealistic worlds in real time from a text prompt or image. Unlike a video generator, it simulates a place you can move through and change while the simulation runs.
What is new in Genie 4 compared with Genie 3?
Genie 4 adds native audio, including music, ambient sound, and interactive dialogue, and it holds a scene together for about two minutes, roughly double Genie 3’s one minute. It also improves physics and human movement and lets you edit a running scene with plain-language prompts.
Can I use Genie 4 right now?
Not yet, unless you are one of the trusted testers in the research preview. Wider access is planned for Google AI Ultra subscribers in Project Genie and early access partners of the Gemini API and Google AI Studio, and a waitlist is open.
How much does Genie 4 cost?
Google has not announced a standalone price for Genie 4. Access is expected to come through existing routes first, namely Google AI Ultra and the Gemini API, so any price will depend on those plans rather than a separate Genie 4 subscription.
What will Genie 4 be used for?
DeepMind points to training environments for robots and agents, plus new interactive experiences in gaming, learning, and film previsualisation. In practice, its first value is likely as a simulator where AI systems can practise safely before acting in the real world.
