Genie 4 is Google DeepMind's newest real-time world model, announced on October 27, 2026. It generates a world you can move through with keyboard and mouse, adds sound to it (music, ambient noise and characters who talk back), and lets you change the scene mid-session by typing, for example "make it rain" or "add a dog". It is a research preview for trusted testers today, with Google AI Ultra subscribers (through Project Genie) and Gemini API early access partners next in line.
Genie 4 is the successor to Genie 3 and belongs to the world model category: AI that learns how an environment looks, sounds and behaves well enough to keep simulating it while you act inside it. Google DeepMind calls it "our most realistic and interactive world model yet" (DeepMind).
| Genie 4 | |
|---|---|
| Maker | Google DeepMind |
| Announced | October 27, 2026 |
| Status | Research preview for trusted testers |
| Next to get access | Google AI Ultra subscribers in Project Genie; early access partners of the Gemini API and Google AI Studio ("in the coming weeks", per TechCrunch) |
| Inputs | Keyboard and mouse to move; text prompts to steer; an image can set the first frame |
| Output | Real-time video with natively generated audio |
| Consistency | Scenes stay consistent for up to about two minutes, double Genie 3, per TechCrunch; Google says sessions support "a few minutes" of interaction |
| Built with | Gemini Omni for realistic appearance and behavior; Gemini 3.8 Flash as an in-the-loop planning agent |
| Watermark | SynthID on all generated video and audio |
| Not published | Resolution, frame rate, latency, price, regions |
Sources: Google DeepMind announcement, Genie model page, TechCrunch.
1. Native audio. DeepMind describes Genie 4 as its first real-time world model to generate audio natively. Wind moves through trees, waves break on a beach, and characters hold dialogue that adapts as you interact. Genie 3 was silent.
2. Steering with plain language. While the world is running you can type a change: different weather, a new camera angle, an extra person or object. Gemini rewrites your prompt to fit the scene and blends between inputs so the world does not jump.
3. Gemini as a director. A smaller Gemini model (Gemini 3.8 Flash) watches the scene, breaks a goal into steps, checks incoming frames to see whether each step happened, and revises its plan if you intervene. DeepMind calls this a "hybrid world model": Gemini plans the story, Genie renders how it looks and sounds.
4. Longer, more physical worlds. TechCrunch reports scenes stay consistent for up to two minutes, about double Genie 3. DeepMind shows collisions, flowing water, wind and characters who move and react naturally.
5. Photos as a starting point. A reference image can be the first frame of a world. DeepMind's example turns a single archival photo into an explorable scene.
| Genie 3 (Project Genie) | Genie 4 | |
|---|---|---|
| Announced | August 2025; Project Genie opened January 29, 2026 | October 27, 2026 |
| Audio | None | Native: music, ambience, character dialogue |
| Change the world mid-session | Text-triggered world events | Free-form edits (weather, camera, people, objects), adapted to the scene by Gemini |
| Consistent scene length | About 60 seconds (Project Genie sessions: 60 s, 720p, 24 fps) | Up to about 2 minutes (TechCrunch) |
| Reasoning layer | World model alone | Gemini plans steps and checks frames |
| Public access | Project Genie, US Google AI Ultra, 18+ | Trusted testers now; Ultra and Gemini API partners next |
Genie 3 specs are from Google's Project Genie launch; Genie 4 resolution and frame rate have not been published, so do not assume they match.
As of October 28, 2026, you cannot open Genie 4 yourself unless Google has made you a trusted tester. The published path is:
Step-by-step details and what to try in the meantime are in our post How to try Genie 4.
DeepMind lists what its trusted testers did with the preview:
Google lists these openly (model page):
Every release side by side, with dates and access rules, is in the 2026 real-time AI video and world model tracker.
Genie 4 is gated, but the experience it points to, video that is generated while you act instead of rendered in advance, is something you can use today in a browser. LiveGen applies it to your own camera: type a prompt in Freestyle and watch your live feed change, or summon a character into your room. No install and no waitlist.
Google DeepMind announced Genie 4 on October 27, 2026, as a research preview for trusted testers. Wider access through Project Genie and the Gemini API was described as coming next, without a fixed date.
No. There is no free tier. Public access is planned through Project Genie, which is part of the $200 per month Google AI Ultra plan, and through Gemini API early access. Google has not published Genie 4 pricing.
Only if you are a trusted tester. Everyone else can join the Project Genie waitlist linked from DeepMind's announcement.
Genie 4 adds native audio, lets you change the world mid-session with text, uses Gemini to plan and check what happens, and keeps a scene consistent for about two minutes instead of about one.
Not publicly. Early access partners of the Gemini API and Google AI Studio are among the first groups to get it. No public endpoint or price has been announced.
Yes. It generates music, ambient sound and character dialogue natively, in real time, alongside the video.
DeepMind shows a reference image used as the first frame, including an archival photo turned into a simulation. Whether that is available to all users at launch has not been stated.
Google says sessions support a few minutes of continuous interaction. TechCrunch reports scenes stay consistent for up to two minutes.
No. A game engine runs hand-built 3D assets with programmed rules. Genie 4 generates every frame from what it has learned, so it can drift, and Google positions it for prototyping, research and simulation.
Google has not said. Project Genie on Genie 3 launched for US users aged 18 and over.
Yes. All video and audio Genie 4 generates carries Google's SynthID watermark.
Open your camera and play any character live — free to start.
Start generating free