Google DeepMind announced Genie 4 on 27 October 2026 — a new world model for unprecedented realism and improved real-time interactive simulation. The DeepMind post is bylined Raia Hadsell, Diego Rivas, and Hang Qi. Genie 4 is natively multimodal: it can simultaneously generate music-like audio, ambient sound, and interactive dialogue with the world — DeepMind’s first real-time world model to generate audio natively.
Built on the Gemini family: grounding and coherence
Genie 4 incorporates and extends the Gemini family: Gemini Omni anchors simulations in real-world appearance and behavior, while other Gemini models supply the complex reasoning that helps keep a simulation coherent and continuous. Users can set an event in motion; Genie 4 simulates what happens next — how the world responds, how it sounds, and how people move, react, and speak.
Keyboard/camera steering and audio waves merging into an explorable scene — real-time steering and native audio metaphor (editorial illustration; no text or logos).
Keyboard/camera steering and audio waves merging into an explorable scene — real-time steering and native audio metaphor (editorial illustration; no text or logos).
Real-time steering: natural language and keyboard
In official demos, users navigate and control the camera with a standard keyboard and steer the evolving scene with text prompts — changing weather, moving the camera, or introducing a person or object — while Gemini contextually adapts prompts to the surrounding scene. Between inputs, Gemini also monitors and steers the scene so the world progresses cohesively. Realism for physics and human action advances as well: Genie 4 can simulate scenarios such as aerodynamics and collisions; characters move naturally, respond organically, and can be steered in real time.
Native audio and dialogue; Gemini 3.8 Flash in the loop
Genie 4 is DeepMind’s first real-time world model to generate audio natively: every world carries ambient sound of its setting, and interactive dialogue adapts on the fly. To model complex multi-step events in real time, Gemini 3.8 Flash serves as a guiding agent-in-the-loop: it observes the scene, breaks goals into steps, and reviews frames to track progress; when the world or user shifts, it adapts the plan in the moment. DeepMind is also experimenting with Gemini 4 Argon, Genie 4, and Gemini Robotics — for example simulating Apptronik’s Apollo humanoid playing checkers, with Argon decomposing the task into real-time captioned steps for Genie to render.
Trusted testers, availability, and safety
A research preview is rolling out to trusted testers across creative industries, journalism, education, gaming, and engineering — for pre-visualization, archival photo→simulation, game concept prototypes, and enterprise zero-to-one scenarios. It is coming to Google AI Ultra subscribers in Project Genie and early-access partners of the Gemini API and Google AI Studio; a waitlist for Genie 4 in Project Genie is open.
Safety uses strict guardrails consistent with other models; internal safety evaluations preceded the preview; all video and audio generated by Genie 4 include SynthID watermarking. Limitations are stated plainly: Genie 4 remains early and experimental; visual detail and object permanence can drift over longer sessions or after complex edits; complex physics, real-time audio accuracy, legible text, spatial accuracy, and long-term consistency are active research areas. Looking ahead, hybrid world models pair Gemini’s structure and narrative planning with Genie’s pixel-level look, sound, and response.
Primary sources:
- Google DeepMind, “Genie 4: A new generation of world simulation,” 2026-10-27, https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/
- Google, “We’re announcing Genie 4…,” 2026-10-27, https://blog.google/innovation-and-ai/models-and-research/google-deepmind/introducing-genie-4/
