DeepMind unveils Genie 4: real-time world simulation with native audio and dialogue
Google DeepMind announced Genie 4, a real-time interactive world model it says improves realism while natively generating ambient sound, music, and character dialogue, with Gemini models helping keep scenes coherent. It is rolling out as a research preview to trusted testers, with later access planned for Google AI Ultra (Project Genie) and early Gemini API / AI Studio partners. DeepMind notes visual detail and object permanence can still drift in longer sessions.
On October 27, 2026, Google DeepMind announced Genie 4 as a new world model aimed at more realistic real-time interactive simulation. DeepMind says Genie 4 is natively multimodal, generating audio-like music, ambient noise, and interactive dialogue alongside the world, and builds on Gemini: Gemini Omni anchors real-world appearance and behavior while other Gemini models supply reasoning to keep simulations coherent. Users can steer with natural language—changing weather, camera, or introducing people and objects—while Gemini monitors and guides the scene between inputs.
DeepMind highlights gains in physics and human-action realism, native audio including on-the-fly dialogue, and multi-step event modeling with Gemini. It describes Gemini 3.8 Flash as an in-the-loop guiding agent that observes frames, breaks goals into steps, and verifies progress, adapting when the world or user changes. Experiments with Gemini 4 Argon and Gemini Robotics interleave Argon-planned text captions into Genie 4 to simulate Apptronik’s Apollo humanoid playing checkers without a physical robot.
Availability: DeepMind says it has started a research preview for trusted testers before consumer and developer release, beginning with Google AI Ultra subscribers in Project Genie and early-access partners of the Gemini API and Google AI Studio, with a Project Genie waitlist. Safety guardrails are described as consistent with other models, with internal evaluations before the preview. Limits (vendor): Genie 4 remains early research; short sessions can stay coherent, but visual details and object permanence may drift over longer runs, complex prompt edits, or navigation; complex physics, real-time audio accuracy, legible text, spatial accuracy, and long-term consistency remain active research areas.
Sources
- Genie 4: A new generation of world simulation — Google DeepMind
- Google Deepmind's Genie 4 puts a modern spin on agent designs that predate LLMs — The Decoder