Google DeepMind announced Genie 4 on October 27, 2026, its new world model for real-time interactive simulation, and the first in the Genie line that generates sound. In DeepMind’s words it is “our new world model, which unlocks unprecedented realism and improves real-time interactive simulation.” You describe a world, move through it with the keyboard, and change it with text prompts while it renders.
It is a research preview. Only trusted testers have it today; Google AI Ultra subscribers and Gemini API partners are next, with no date and no price. This page is read from DeepMind’s announcement, Google’s blog post and the Genie model page, all read on launch day.
On this page · 6 sectionsOpen
Google DeepMind announced Genie 4 on October 27, 2026: a world model that generates explorable scenes in real time, steered with keyboard controls and text prompts.
It is Google’s first real-time world model with native audio: ambient sound, music and characters who speak and react to what you do.
Gemini 3.8 Flash runs in the loop as a guiding agent, breaking a goal into steps, watching the frames and checking each step succeeded.
Google gives no resolution, frame rate or session length. Genie 3 stated 720p, 24 frames per second and a few minutes of consistency.
Only trusted testers have it today. Google AI Ultra subscribers in Project Genie and Gemini API early-access partners are next, with no date or price.
§ 01What Genie 4 does
DeepMind names four capabilities:
- Steer the world live. Change the weather, move the camera or add a person with a text prompt, and Gemini adapts the prompt to the scene. Between your inputs, Gemini keeps the world moving coherently.
- Physics and people. More realistic collisions and aerodynamics, and characters who “move naturally, respond organically to their environments”, in DeepMind’s description.
- Sound. “Genie 4 is our first real-time world model to generate audio natively.” Every world gets ambient sound, and characters hold dialogue that reacts to what you do.
- Gemini in the loop. For multi-step events, “Gemini 3.8 Flash serves as a guiding agent in the loop”: it breaks a goal into steps, watches incoming frames, and checks that each step succeeded before the next.
The last one is the agent story. DeepMind also shows Gemini 4 Argon planning a task step by step for Genie to render: Apptronik’s Apollo humanoid playing checkers in a simulated room. That is the robotics use case, testing an agent in an imagined world before it touches a real one.
§ 02Genie 4 against Genie 3
Genie 3, announced in August 2025, came with numbers: it could generate worlds “at 24 frames per second, retaining consistency for a few minutes at a resolution of 720p.” Genie 4’s announcement gives none of those figures.
| Genie 3 (August 2025) | Genie 4 (October 27, 2026) | |
|---|---|---|
| Resolution | 720p | Not stated |
| Frame rate | 24 frames per second | Not stated |
| How long a world holds together | A few minutes | “Short sessions”, with drift over time |
| Native audio | No | Yes: ambient sound, music, dialogue |
| Planning model in the loop | None stated | Gemini 3.8 Flash; Gemini 4 Argon in a robotics demo |
Read the blank cells as blanks, not as Genie 3’s numbers carried over. DeepMind is candid about limits: “Complex physical interactions, real-time audio accuracy, rendering legible text, spatial accuracy, and long-term consistency are active areas of ongoing research.”
§ 03What changed on November 18
Google opened Genie 4 to the public side of Project Genie on November 18, 2026. Its post says: “Eligible Google AI Ultra subscribers can try the new Project Genie experience starting today, and Google AI Pro members can join the waitlist on the same page.”
Four things arrived together. Genie 4 can be edited mid-session by typing a new prompt while you move through the world. It adds sound: “For the first time, Genie 4 pairs these visuals with real-time native audio that adapts as the scene unfolds, from ambient soundscapes to character dialogue.” A Project Genie agent builds a world from a conversation, and a community gallery lets people publish and remix worlds.
Genie 3 stays in the model picker for longer runs, with a longer limit per generation: “Explore for up to 3 minutes per generation (up from 60 seconds), or link multiple worlds together for longer journeys.” Google still lists open problems: “camera latency, intricate physical interactions, real-time audio precision, and long-term scene permanence”.
| Change | Before | After |
|---|---|---|
| Genie 4 access | Trusted testers | Eligible Google AI Ultra subscribers |
| Google AI Pro members | No access | Waitlist |
| Genie 3 time per generation | 60 seconds | Up to 3 minutes |
| Building a world | One text or image prompt | A conversation with the Project Genie agent |
| Sharing | None | Community gallery to publish and remix |
§ 04Who can use it, and when
| Group | Status |
|---|---|
| Trusted testers | Research preview has started |
| Google AI Ultra subscribers, in Project Genie | Open to eligible subscribers since November 18, 2026 |
| Google AI Pro members | Waitlist on the Project Genie page |
| Early-access partners, Gemini API and Google AI Studio | Coming soon (October 27 post; not in the November 18 post) |
| Price | No separate price in Google’s November 18 post |
Google says testers in film, journalism, education, games and engineering are using it as a “draft mode”: filmmakers running several takes of a scene live, photojournalists turning one archival photo into a walkable scene, and game developers testing a mechanic before writing engine code. It also says all video and audio generated by Genie 4 will include SynthID, its watermark.
Google DeepMind posted the launch on X at 16:10 UTC.
§ 05What it means for teams that make video
Genie 4 is a different kind of tool from a video generator. A video model gives you a finished clip you can edit and ship; a world model gives you a scene you steer while it runs, which is why the trusted testers use it for previsualization rather than final output. Since November 18, eligible Google AI Ultra subscribers can try both uses in Project Genie.
CellCog does not route any work to Genie. Our AI employees make finished video with a reference-to-video model that generates dialogue and sound natively, and if Genie 4 opens a usable API with a fit for that work, we will test it the way we test every model.
§ 06What to watch
-
Source of the November 18 update: Google, Genie 4 is now available, November 18, 2026.
-
The wider rollout to Google AI Pro members on the waitlist, and whether a separate price appears.
- Numbers. Resolution, frame rate and session length, from Google or from the first outside testers.
- The Gemini API. Whether developers get programmatic access, which is what agent training and robotics teams need.
Q1When was Genie 4 announced?
On October 27, 2026, in posts on the Google DeepMind blog and Google’s main blog, and on Google DeepMind’s X account at 16:10 UTC.
Q2Does Genie 4 generate sound?
Yes. Google says it is its first real-time world model to generate audio natively, including ambient sound, music and interactive dialogue. It also lists real-time audio accuracy as an open research problem.
Q3How long does a Genie 4 world stay consistent?
Google does not give a number. It says Genie 4 holds coherent environments over short sessions, and that details can drift as a simulation runs longer, after several complex prompt edits, or while you move through the scene.
Q4How much does Genie 4 cost?
Google has not announced a price. The first public route is Project Genie for Google AI Ultra subscribers, which is listed as coming soon.
Q5Is Genie 4 output watermarked?
Yes. Google says all video and audio generated by Genie 4 will carry SynthID, its imperceptible watermark.
