Google DeepMind unveils Genie 4, a real-time world model with audio
Google DeepMind's Genie 4 research preview generates interactive worlds with native sound and dialogue that you can re-steer mid-session with text prompts.
· 3 min read · By Efe Anidi
Google DeepMind announced Genie 4 on 27 October 2026. It is a world model that builds interactive environments in real time, and the first in the Genie series to produce its own sound. The announcement is credited to Raia Hadsell, Diego Rivas and Hang Qi. Genie 4 is a research preview for trusted testers, not a public product.
What Genie 4 can do
While a simulation runs, Genie 4 generates ambient sound, music and character dialogue to match what is on screen. Users move through the world and control the camera with the keyboard. They can also type instructions mid-session: make it rain, switch the viewpoint, drop in a person or an object. Gemini rewrites each instruction to fit the current scene, so edits do not break the world.
DeepMind says the “most dramatic leap” is in how Genie 4 simulates people. Characters move more like humans, respond to what happens around them, and can talk with the user live. The post also claims better physics, citing aerodynamics and collisions as examples, but gives no technical detail. It also gives no resolution, frame rate or session-length figures. TechCrunch reports that a scene can stay consistent for up to two minutes.
How Gemini fits in
DeepMind presents Genie 4 as one part of a system built with the Gemini family. Gemini Omni keeps scenes looking and behaving like the real world, and other Gemini models handle reasoning. For multi-step events, Gemini 3.8 Flash watches the scene, splits a goal into steps, checks whether each one happened, and re-plans if it didn’t. Genie 4 does the rendering.
A separate experiment, linked to Gemini Robotics, pairs Genie 4 with a model DeepMind calls Gemini 4 Argon. Argon plans a robot task and sends the steps to Genie 4 as text captions while the scene renders. The examples include an Apptronik Apollo humanoid playing checkers and a Universal Robots UR5 arm pipetting liquid into a lab plate. In other words, robot behaviour gets rehearsed in simulation before anyone touches hardware. TechCrunch quotes Google research director Tejas Kulkarni, who expects more of these combined language-model-plus-simulator systems, which he calls hybrid world models.
Why it matters for world builders
For anyone prototyping interactive stories, the main change is that you can direct a running world in plain language instead of just walking through it. One prompt can work as a stage direction and a camera move at the same time. That is the pattern Runway is chasing with GWM Worlds 2, and DeepMind now has Gemini doing the planning and consistency checks on top.
DeepMind’s trusted testers have used Genie 4 for quick previsualisation and alternate takes for film, for turning a single archival photo into an explorable scene, and for sketching visual styles and mechanics for interactive experiences. Enterprise partners are testing it for scenario generation.
Availability and limits
Today, only trusted testers can use Genie 4. DeepMind says Google AI Ultra subscribers will get it through Project Genie, along with early-access partners on the Gemini API and Google AI Studio. TechCrunch puts that rollout in the coming weeks. There is a waitlist, but no price.
DeepMind is upfront about weaknesses. Objects and visual details can drift, especially in long sessions, after several edits, or while you move around. Complex physical interactions, accurate real-time audio, readable in-world text, spatial accuracy and long-term consistency are all listed as unsolved. All generated video and audio carries a SynthID watermark, and DeepMind says it ran internal safety evaluations before the preview.
Builders cannot use Genie 4 yet. Still, it is the clearest sign so far that DeepMind sees world models as something you direct in conversation, with a language model planning and checking what the simulator shows.
Questions and answers
What is Genie 4?
Genie 4 is Google DeepMind's newest world model: a system that generates explorable, interactive environments in real time. Announced on 27 October 2026, it is the first Genie model to generate sound natively, including music, ambient audio and spoken dialogue from characters.
Can I use Genie 4 today?
Only if you are a trusted tester. DeepMind says Genie 4 will reach Google AI Ultra subscribers through Project Genie, plus early-access partners on the Gemini API and Google AI Studio, and there is a Project Genie waitlist. No pricing or fixed release date has been published.
How is Genie 4 different from Genie 3?
Genie 4 adds native audio, more photorealistic visuals, more natural characters and free-form text edits to a running scene. TechCrunch reports it can keep a scene consistent for up to two minutes. DeepMind's own post gives no resolution, frame rate or duration figures.
Sources:
