Google DeepMind announced Genie 4 on October 27, 2026, a world model that turns text prompts and keyboard input into interactive, photorealistic environments and, for the first time in the Genie line, generates sound and spoken dialogue along with the picture. The model is rolling out as a research preview to trusted testers, with a waitlist open for Google AI Ultra subscribers in Project Genie and early-access partners of the Gemini API and Google AI Studio.
What Genie 4 does
A world model predicts what happens next in an environment, so that a user can act inside it instead of just watching a clip. DeepMind says you can “set an event in motion” and Genie 4 simulates how the world responds, how it sounds, and how the people in it move, react and speak. Users steer the scene with standard keyboard controls for navigation and the camera, and with natural-language prompts that change the weather, move the camera or add a person or object while the simulation keeps running.
According to the announcement, Genie 4 builds on the Gemini family. A model named Gemini Omni anchors simulations in real-world appearance and behavior, while other Gemini models supply the reasoning needed to keep a simulation coherent from one moment to the next. Between a user’s inputs, Gemini also monitors the scene and steers it so the world keeps progressing in a consistent way.
Native audio and interactive dialogue
DeepMind calls Genie 4 its first real-time world model to generate audio natively. Every world carries the ambient sound of its setting, from wind in trees to waves on a beach. The lab describes human speech as the harder frontier: characters can converse and react to a user’s inputs or to events in the scene, and the dialogue adapts on the fly.
The post also says Genie 4 simulates real-world physics scenarios such as aerodynamics and collisions, and that its most dramatic improvement is in how it simulates people, who are said to move naturally and respond to their surroundings while being steered in real time.
Why it matters
DeepMind frames world models as part of the road to AGI and names two uses: creating training environments for robots and agents, and building new interactive experiences. The first is the more consequential for the industry. Physical AI developers need large amounts of varied, safe simulation, and rivals are working on the same problem; our earlier report on NVIDIA’s Cosmos world-model tools covers one approach aimed at robotics teams. For a short definition of the idea, see our glossary entry on the world model.
Safety, watermarking and limits
DeepMind says Genie 4 has safety guardrails consistent with its other models and went through internal safety evaluations, and that it is being released to trusted testers first. All video and audio the model generates will carry SynthID, the company’s imperceptible watermark for AI-generated content.
The lab is also candid about the limits. Genie 4 keeps environments coherent over short sessions, but visual details and object permanence can drift, especially in longer runs, after several complex prompt edits, or when moving around the scene. Complex physical interactions, real-time audio accuracy, legible text, spatial accuracy and long-term consistency are listed as open research problems. The post does not give pricing, session length, resolution or a general release date, so those remain unknown.
What to watch next
- When Genie 4 reaches Google AI Ultra subscribers in Project Genie and how long sessions can run.
- Whether the Gemini API and Google AI Studio previews expose the model to developers building agent or robot training loops.
- Independent tests of object permanence and physics, which DeepMind itself flags as weak spots.
- Competing world models from other labs and chipmakers.
- How SynthID detection holds up when Genie 4 clips are shared and re-encoded.
Key takeaways
- Genie 4 is DeepMind’s new real-time world model, announced October 27, 2026.
- It generates audio and interactive character dialogue natively, a first for the Genie line.
- It is built on Gemini models, including Gemini Omni for real-world appearance.
- Access starts with trusted testers, then Google AI Ultra and early-access API partners.
- DeepMind says long-session consistency and complex physics are still unsolved.
Frequently asked questions
What is Genie 4?
Genie 4 is Google DeepMind’s latest world model. It generates interactive, photorealistic environments in real time from text prompts and keyboard controls, and adds native audio and character dialogue.
Can I use Genie 4 today?
Not yet in general. It is a research preview for trusted testers, with a waitlist for Google AI Ultra subscribers in Project Genie and early-access partners of the Gemini API and Google AI Studio.
Is Genie 4 output labeled as AI-generated?
Yes. DeepMind says all video and audio produced by Genie 4 includes SynthID, its imperceptible digital watermark for identifying AI-generated content.
