Google DeepMind Unveils Genie 4, an Interactive World Model With Live Audio and Speech

Built on Gemini Omni for real-time world generation

On October 27, 2026, Google DeepMind unveiled Genie 4 as the newest milestone in its Genie line of interactive world models—part of a research portfolio on deepmind.google alongside Gemini Omni, Veo, Lyria, SIMA, and Gemini Robotics. Whereas conventional video generators render a fixed clip from a prompt and stop, a world model continuously computes the next state of an environment as a user explores and acts within it.

Genie 4 is built on Google DeepMind's Gemini Omni multimodal foundation. That underlying architecture allows the system to move past silent, low-resolution physics sandboxes and generate photorealistic, dynamic scenes that respond in real time as the user navigates through them.

Scenes that sound, react, and populate with speaking characters

The defining leap in Genie 4 is multimodal aliveness inside the generated world. In Google DeepMind's official demonstration, exploring a Genie 4 environment is no longer purely visual: the scene generates matching environmental sound on the fly, and the people inside the world move, react to what is happening around them, and speak.

Users can steer and probe the scene through natural instructions while watching characters and surroundings adapt on the spot. That combination of interactive state updates, synchronized acoustics, and human dialogue brings generative world models closer to playable simulations, virtual production sets, and interactive training environments for embodied agents.

• Real-time scene response

  Environments update dynamically as you explore, maintaining visual continuity rather than playing back a pre-rendered video.

• Native environmental sound

  Scenes generate their own audio alongside the visuals as events unfold inside the world.

• Reactive, speaking people

  Characters inside the generated environment move, react to interactions, and speak in real time.

Current research preview status and what remains unannounced

Genie 4 is currently available only as a limited research preview for selected testers. General users cannot yet launch interactive sessions directly in a public browser window, and Google DeepMind has not published an open sign-up form for the tester cohort.

Key commercial details—including a public release date, pricing tiers, supported regions, and full technical documentation of the control interface—have not been announced. For now, anyone tracking Genie 4 can view the official overview and demonstration footage on deepmind.google while Google DeepMind evaluates the system with its preview testers.