Back to LLM Models

Genie 4: Google's interactive world model with native audio

Google DeepMind announced Genie 4 on October 27, 2026, a world model that generates interactive environments in real time from prompts. The changes versus its predecessor, according to the company, are three: native audio (ambient sound and interactive dialogue generated alongside the video), photorealism, and steering of the simulation with natural language while it runs. The launch is a research preview for trusted testers, with broader access starting with Google AI Ultra subscribers through Project Genie and early-access partners of the Gemini API and Google AI Studio.

A world model, not a video model

In DeepMind's framing, a world model simulates what happens next: you set an event in motion and the model generates how the world responds, what it sounds like, and how the people in it move, react and speak. Genie 4 builds on the Gemini family. Gemini Omni anchors the simulations in real-world appearance and behavior, and other Gemini models supply the reasoning that keeps the simulation coherent over time. The company frames this as infrastructure for training robots and agents, and for building interactive experiences.

Real-time steering, physics, people

The post lays out four capability areas:

  • Real-time steering. Natural-language input can change the simulation mid-run: different weather, a moved camera, a new person or object added. Gemini adapts the prompt to the surrounding scene, and between your inputs it monitors and steers the scene to keep the world progressing.
  • Physics and human motion. DeepMind says the model simulates real-world physics scenarios such as aerodynamics and collisions. The biggest jump, the company claims, is in simulating people: characters that move naturally, react to their environment, and can be steered interactively in real time.
  • Native audio. Genie 4 is DeepMind's first real-time world model to generate audio natively. Every world carries its own ambient sound, and the post also shows interactive dialogue, with characters conversing in real time and responding to user input.
  • Multi-step events. Paired with Gemini, the model handles complex, multi-step scenarios instead of single moments.

Gemini in the loop

The multi-step work runs with Gemini 3.8 Flash as a guiding agent in the loop: while Genie 4 renders the world, Gemini watches the scene, breaks the goal into steps, and reviews incoming frames to verify each step succeeded, adapting the plan as the world shifts or the user intervenes. The post also describes experiments with Gemini 4 Argon and Gemini Robotics, where Argon decomposes a user-instructed task into steps that are fed to Genie 4 as text captions during rendering, with a demo of Apptronik's Apollo humanoid playing checkers.

Who is testing it

DeepMind has been running cohorts of trusted testers across creative, journalism, education, gaming and engineering. According to the post, filmmakers use it as a "draft mode" for pre-visualization, photojournalists turn a single archival photograph into a live simulation, game developers go from an idea to an explorable prototype before writing engine code, and enterprise partners are evaluating it for "zero-to-one" scenario generation.

Availability and limits

The model is a research preview and, in DeepMind's own words, an early and experimental one. Access is limited to trusted testers, with a waitlist for Project Genie, and no pricing has been published. All generated video and audio will carry a SynthID watermark.

The stated limitations include visual details and object permanence drifting over time, especially in long sessions or after multiple complex prompt edits, plus ongoing work on text rendering, spatial accuracy, real-time audio accuracy and long-term consistency. If you are weighing this as a working tool, the practical read today is a watch item, not a buy.

Source: deepmind.google

More news

Back to LLM Models

Have something in mind?

Start a conversation