Google DeepMind unveils Genie 4, a world model that also generates its own sound

Google DeepMind announces Genie 4 as a research preview

Google DeepMind announced its new world model, Genie 4, on October 27. World models are AI systems that generate, frame by frame, what an environment looks like and how it responds to a user’s actions. Genie 4 builds a real-time, interactive world from a written prompt. Users move through it and control the camera with standard keyboard keys.

The company calls Genie 4 “our most realistic and interactive world model yet.” DeepMind says understanding and simulating the world is a crucial part of building AGI. It says the model could help build safe training environments for robots and AI agents.

Google DeepMind@GoogleDeepMind

This is Genie 4, our most realistic and interactive world model yet. 🌐

Explore dynamic environments in real time to see how the scene responds, sounds, and how the people in it move, react, and speak.

Gemini Omni handles the look, other Gemini models keep it coherent

Genie 4 builds on capabilities from the Gemini family. According to DeepMind, Gemini Omni anchors simulations in real-world appearance and behavior. Other Gemini models provide the reasoning needed to keep a scene coherent and continuous.

The company says the model can simulate physics such as aerodynamics and collisions. It calls its handling of people the most dramatic leap: characters move naturally, respond to their surroundings and can be steered in real time. With a written prompt, users can change the weather, move the camera, or add a new person or object. Gemini adapts the prompt to fit the rest of the scene.

TechCrunch, citing a press briefing, reports that a scene can stay consistent for up to two minutes, double the limit of Genie 3. That figure does not appear in DeepMind’s own announcement.

Google DeepMind@GoogleDeepMind

Here’s what’s new with Genie 4:
🔵 Photorealism: Built on Gemini Omni for lifelike detail across spaces and subjects
🔵 Advanced physics: Aerodynamics, collisions, and people who move naturally
🔵 Real-time controls: Prompt to change the scene, add elements, or interact with characters
🔵 Native audio: Ambient sound, music, background noise, and interactive dialogue
🔵 Gemini reasoning: Plans multi-step events to maintain coherence as scenes evolve

These capabilities are a step toward developing safe virtual environments to see how agents and robots navigate their surroundings.

We’re exploring early use cases with trusted testers first, before bringing the model to Google AI Ultra subscribers in Project Genie and early access partners in our API and @GoogleAIStudio soon.

Find out more → goo.gle/48idm3b

Sound, music and dialogue are generated with the visuals

DeepMind says Genie 4 is its first real-time world model to generate audio natively. Every world carries the ambient sound of its setting, from wind in the trees to waves on a beach. Music and character dialogue are generated at the same time as the visuals.

The company calls human speech the real frontier. In Genie 4, characters can talk back in real time in response to the user’s prompts and to events in the world. In a dragon race example DeepMind published, wingbeats, wind and a distant volcanic rumble were generated along with the visuals, as described in the prompt. TechCrunch notes that Runway’s GWM Worlds 2, released a couple of months earlier, also generates audio.

Gemini 3.8 Flash watches the scene and steers it step by step

According to DeepMind, Gemini 3.8 Flash acts as a “guiding agent in the loop” during real-time generation. As Genie 4 renders the world, Gemini continuously observes the scene and breaks the goal into steps. It reviews incoming frames to check whether each step succeeded. When the world shifts or the user steps in, it adapts its plan on the spot.

DeepMind is also experimenting with Gemini 4 Argon alongside Genie 4 and its Gemini Robotics work. In an example the company shared, Argon breaks down a game of checkers played by Apptronik’s Apollo humanoid into steps. Those steps are fed to Genie 4 as text captions while the simulation runs. TechCrunch says this could let teams test robot scenarios before trying them in the real world. Google research director Tejas Kulkarni described the goal as “hybrid world models,” with Gemini writing code and text.

Trusted testers get it first

For now, Genie 4 is a research preview for trusted testers. DeepMind says it has been working with tester groups from creative industries, journalism, education, gaming and engineering. According to the company, filmmakers are trying it as a fast “draft mode” for previsualization. Photojournalists are turning single archival photos into simulations, and game developers are building prototypes before writing engine code.

The model is coming soon to Google AI Ultra subscribers in Project Genie and to early access partners in the Gemini API and Google AI Studio. TechCrunch reports that this rollout is expected in the coming weeks; no exact date was given. Project Genie opened on Jan. 29, 2026, as a Genie 3 prototype for AI Ultra subscribers in the U.S. In that version, each generation was capped at 60 seconds.

DeepMind spells out the limits

DeepMind describes Genie 4 as an early, experimental model. The company says it stays coherent over short sessions. But visual details and object permanence can drift as a simulation runs longer, after several complex prompt edits, or while moving through the scene. Complex physical interactions, real-time audio accuracy, legible text, spatial accuracy and long-term consistency remain active areas of research.

All video and audio generated by Genie 4 carries a SynthID watermark, which helps identify AI-generated content. DeepMind says it ran internal safety evaluations before the preview. The announcement gives no resolution, frame rate or latency figures, and no measured comparison with Genie 3. Genie 3 was announced in August 2025 with 720p resolution at 24 frames per second.