Google DeepMind has announced Genie 4, the next version of its world model, which generates an explorable world in real time as you move through it. The DeepMind post and Google’s announcement focus on two new abilities: sound, and steering with words.
Sound and talking characters
Genie 4 is “natively multimodal”. Each world generates audio as it goes, “including ambient sounds, music, and interactive dialogue”. Google calls speech “the true frontier”: characters “converse and react to your inputs or events in the world, all in real-time”.
Steering a world while it runs
You can change a scene mid-session with natural language, for example “moving the camera, changing the weather, or introducing new characters”. Under the hood, Google says, Gemini Omni anchors the look and behaviour of the world, while Gemini 3.8 Flash acts as “a guiding agent in the loop”: it watches the scene, breaks a goal into steps and checks each new frame to see whether a step worked.
DeepMind also shows a robotics experiment: Gemini 4 Argon breaks a task into steps that Genie 4 renders, in a demo of Apptronik’s Apollo humanoid playing checkers. That is a simulation for planning research, not a robot product.
Who can use it
Not most people, yet. Genie 4 is “a research preview for trusted testers”. Google says it is “coming soon” to Google AI Ultra subscribers in Project Genie and to early-access partners in the Gemini API and Google AI Studio, with a waitlist. There are no dates or prices.
All video and audio Genie 4 generates will carry SynthID watermarks. Google lists the limits plainly: details and object permanence can “shift or drift over time”, and complex physics, audio accuracy, legible text, spatial accuracy and long-term consistency are still open problems.
Update, 19 November 2026: Genie 4 is open to AI Ultra subscribers
Google rolled Genie 4 out on 18 November to eligible Google AI Ultra subscribers, inside a rebuilt Project Genie. AI Pro members can join a waitlist. The post does not mention Gemini API or AI Studio access, so the developer route described above is still unannounced.
What Ultra subscribers get:
- Genie 4 with mid-session text prompts that change the world as you move through it, plus real-time native audio.
- Genie 3 upgrades, still in the model picker: up to three minutes per generation (up from 60 seconds), chained worlds, and worlds seeded from 360-degree panoramas.
- A Project Genie agent that builds a world from a conversation, using Google’s other models for the visuals and a background score, with Gemini able to join a session.
- A community gallery where worlds can be published and remixed.
SynthID watermarks still apply to all video and audio. Google again lists camera latency, physical interactions, audio precision and long-term scene permanence as open problems.
Update, 20 November 2026: Genie 4 is generally available in the Gemini API
The Gemini API release notes list Genie 4 (models/genie-4) as generally available on 19 November, in the Interactions API (v1beta). That answers the open question from yesterday’s update: developers can now build on it directly.
Per the release notes:
- Output. Real-time, steerable 24 fps 16:9 video with synchronized 48 kHz stereo audio, streamed over WebRTC or WebSockets.
- Input. Seed a world from text, a 16:9 reference image, or both, then steer it with WASD or arrow keys and text events during the session.
- Price. Paid tier only: $1.50 per million input tokens, $9.00 per million text output tokens and $17.50 per million video output tokens, which Google puts at about $0.10 per second of video, audio included.
At roughly $0.10 a second, an hour of live world streaming costs about $360 in video output alone, so budget for session length before you put Genie 4 in front of users.

The Campfire
CommentsThanks. Your comment is waiting for the desk. Once it is approved it appears here within about a minute.
Add a comment