Google DeepMind · Genie 3 on August 5, 2025 · Genie 4 on October 27, 2026
DeepMind Genie 4: the step from Genie 3, not a live world
DeepMind Genie 4 is Google DeepMind's name for the world model announced on October 27, 2026. Genie 3, on August 5, 2025, was the first of their world models they described as real-time. The form on this page runs Veo 3.1 Fast. It returns one clip. The Gemini-in-the-loop write-up stays on the Gemini Genie 4 page.
This generator runs Veo 3.1 Fast for 60 credits a clip. It is not DeepMind Genie 4, and it does not keep a world you can steer.
Ratio
Generated Videos
No videos yet
Fill out the form and click Generate Video to create your first AI video.
Tips: Use detailed descriptions for better results
- Made by
- Google DeepMind
- Released
- October 27, 2026
- Run on Nano AI
- Veo 3.1 Fast · from 60 credits per video
- Page updated
- November 2, 2026
Source: deepmind.google
What changed from Genie 3 to DeepMind Genie 4
Google DeepMind's August 5, 2025 post, by Jack Parker-Holder and Shlomi Fruchter, announces Genie 3. It says a text prompt can become a world you navigate in real time at 24 frames per second, consistent for a few minutes, at 720p. It calls Genie 3 the first of their world models to allow real-time interaction, and an improvement on Genie 2. It also separates that work from video models, naming Veo 2 and Veo 3 as video generation. Promptable world events are the text edits that change weather or add an object while you navigate. The post lists limits: a narrow action space, other agents, real places that are not geographically exact, text that is clear mainly when it was in the description, and a few minutes rather than hours. Access then was a limited research preview.
The October 27, 2026 post, by Raia Hadsell, Diego Rivas, and Hang Qi, announces Genie 4. It says the model is natively multimodal, so music, ambient noise, and dialogue can be generated with the world, and it calls Genie 4 their first real-time world model to generate audio natively. It says a research preview has started for trusted testers, with a later step for Google AI Ultra subscribers in Project Genie and early-access partners of the Gemini API and Google AI Studio. It says generated video and audio include SynthID. Limits they repeat: details and object permanence can drift as a session runs, after complex edits, or while you move through the scene. Physics, audio accuracy, legible text, spatial accuracy, and long consistency are still research. This page does not repeat the Gemini 3.8 Flash agent loop. That belongs to the Gemini Genie 4 page, which answers a different search.
Veo 3.1 Fast on this site is 60 credits a clip. It is a text-to-video model this site has priced. It is not the Veo 3 named in the Genie 3 post, and it is not a session at 24 frames per second and 720p that you steer. Eligible new accounts get 64 credits, which covers one clip. The file ends when it finishes.
DeepMind Genie 4 at a glance
- Genie 3 post
- August 5, 2025. Jack Parker-Holder and Shlomi Fruchter
- Genie 3, their words
- Real time, 24 frames per second, a few minutes, 720p
- Against Genie 2
- First of their world models they call real-time
- Promptable events
- Text can change weather or add an object, in the Genie 3 post
- Genie 4 post
- October 27, 2026. Raia Hadsell, Diego Rivas, and Hang Qi
- Audio
- Native audio. They call it the first real-time world model of theirs with it
- Access
- Trusted-tester preview. Ultra in Project Genie and API partners described as later
- Watermark
- SynthID on video and audio, their words
- Companion page
- Gemini Genie 4. The Gemini loop is not repeated here
- On this page
- Veo 3.1 Fast, 60 credits a clip. Not a Genie session
How to read the Genie 3 to Genie 4 step
The publisher is DeepMind
Both posts are on deepmind.google. Gemini Genie 4 is a different search for the same October 27 announcement. This page is the naming and the step from Genie 3.
24 frames per second is Genie 3
The August 5, 2025 post pairs 24 frames per second with 720p and a few minutes. This page does not claim Genie 4 published a new frame rate. The Veo clip is not that session.
Audio is the Genie 4 addition they stress
The October 27 post says music, ambient sound, and dialogue are generated with the world, and that SynthID marks the video and the audio. That claim is about Genie 4.
One clip is not navigation
Veo 3.1 Fast costs 60 credits and stops. It does not offer promptable world events, and 64 credits cover one clip.
Genie 3, Genie 4, and the clip on this page
Genie 3 is the August 5, 2025 post. Genie 4 is the October 27, 2026 post. Both were read on 2026-11-02. The Veo column is what Nano AI runs. The Gemini agent loop is on the other page.
| Genie 3 | DeepMind Genie 4 | Veo 3.1 Fast here | |
|---|---|---|---|
| Date on the post | August 5, 2025 | October 27, 2026 | Not a DeepMind release |
| What they say it is | A world you navigate. First real-time versus Genie 2 | A newer world model with native audio | One generated clip |
| Motion they print | 24 frames per second, 720p, a few minutes | Real-time steering. No new frame rate copied | A clip. Not a 720p session you steer |
| Sound | Not described as native audio | Music, ambient sound, and dialogue, with SynthID | A clip may include sound. It is not a Genie session |
| Who can use it | Limited research preview, in that post | Trusted testers. Wider access described as later | The form above, 60 credits |
| Not on this button | No promptable world events | No Project Genie waitlist | This is the button |
How to make a short clip on this page
- 1
Stay on Veo 3.1 Fast
The generator is set to Veo 3.1 Fast. One clip is 60 credits. Switching models changes the price. It does not start Genie 3 or Genie 4.
- 2
Pick an empty scene
Use one of the scene buttons or edit the prompt. Describe a place and a camera move. Do not write a film title or a real person's name.
- 3
Generate one clip
Sign in. The cost is shown before the clip starts. Eligible new accounts get 64 credits, which covers one clip. The file will not continue as a world you can walk through.
Scenes for one clip
These prompts are for Veo 3.1 Fast. They do not start a Genie 3 or Genie 4 session.
Glass greenhouse
An empty glass greenhouse at dawn, wet leaves, one open aisle, the camera glides forward, soft drips, no people, no readable text.
Empty cable car
An empty cable car moves above a pine valley, morning mist, the camera looks out the window, no people, no readable text.
Stone steps
Wet stone steps beside a quiet canal, one moored boat, cool light, the camera holds, water against stone, no people, no readable text.
Paper kite
A plain paper kite rests on dry grass, a light wind lifts one corner, wide sky, no people, no logos, no readable text.
FAQ
DeepMind Genie 4 FAQ
Dates and limits from the two DeepMind posts, as read on 2026-11-02, and what this website runs.
Is DeepMind Genie 4 the same page as Gemini Genie 4?
No. Both names point at the October 27, 2026 announcement. This page is the DeepMind name and the step from Genie 3 on August 5, 2025. The Gemini Genie 4 page covers Gemini's role inside the simulation and is not copied here.
What did Genie 3 publish that Genie 4's post does not repeat as a spec?
The August 5, 2025 post says 24 frames per second, 720p, and a few minutes, and it names promptable world events. This page does not treat those numbers as a Genie 4 spec, because the October 27 post does not restate them as new figures.
What does the Genie 4 post add?
Native audio, including music, ambient sound, and dialogue, plus SynthID on the video and the audio. Access is a trusted-tester preview, with Google AI Ultra in Project Genie and early API partners described as later.
Can I steer a world on this page?
No. Veo 3.1 Fast returns one clip for 60 credits. It is not a real-time session, and it is not the Veo 3 named in the Genie 3 post.
What limits do the posts share?
Both say a session is short, real places are not exact, and on-screen text is unreliable unless it was in the description. Genie 4 also says details can drift after edits or while you move through the scene.
Is Nano AI affiliated with Google or DeepMind?
No. Nano AI is not affiliated with Google, Google DeepMind, Gemini, Genie, or Project Genie. The generator runs Veo 3.1 Fast.
More from Nano AI
Related pages
Nano AI is an independent service and is not affiliated with Google, Google DeepMind, Gemini, Genie, or Project Genie. Genie 3 is from the August 5, 2025 DeepMind post. DeepMind Genie 4 is from the October 27, 2026 post. Both were read on 2026-11-02. The Gemini loop is on the Gemini Genie 4 page. This generator runs Veo 3.1 Fast.