Google Unveils Genie 4: Real-Time, Audio-Enabled World Model Pushes Boundaries of Interactive Simulation

From Genie 3 to Genie 4: Google’s Latest Leap in World Modeling

Google DeepMind launched Genie 4, its newest generative world model, in a research preview on Tuesday. The system produces the most photorealistic and interactive virtual environments seen to date, letting users type prompts during a simulation to alter the scene on the fly. In addition, Genie 4 generates synchronized audio that mirrors the simulated world, encompassing music, ambient sounds, and interactive dialogue.

The debut of Genie 4 follows a similar release by AI video startup Runway, which introduced GWM Worlds 2 a couple of months ago. That model also responds dynamically to user inputs and includes audio generation. Meanwhile, the broader world‑model sector is consolidating: AMD is acquiring Fei‑Fei Li’s World Labs, Amazon is in talks to purchase Decart, and Rivian has bought Third Dimension to bring simulation capabilities in‑house for autonomous‑driving testing.

Genie 4 builds on its predecessor, Genie 3, which debuted as a research preview just over a year ago. While Genie 3 already delivered higher photorealism and introduced audio, Genie 4 adds freeform edits such as changing weather or camera angle in real time and inserting new characters or objects. It can sustain a coherent scene for up to two minutes—double the limit of Genie 3—and offers improved physics and more human‑like avatars.

Gemini, Google’s large language model, is now tightly coupled with Genie 4, serving as a director that outlines steps and validates frames while Genie renders the visuals. In a demonstration linked to Gemini Robotics, a Gemini 4 Argon instance decomposed a task into sub‑steps that Genie 4 simulated instantly. The experiment illustrates how robot scenarios could be rehearsed in simulation before real‑world execution.

“What we’ll increasingly see in the future are these hybrid world models, where you have a combination of Gemini writing code and text, and Genie ingesting that in real time and simulating reality at different levels of abstraction,” Tejas Kulkarni, research director at Google, said during a briefing.

Potential commercial applications echo those of earlier world models: gaming, interactive education, and robotics. One imaginative use case envisions stepping into a painting of a historical event and conversing with the figures depicted. Trusted testers—those granted early access to Google’s cutting‑edge products—have also suggested that the real‑time nature of Genie 4 could streamline film prototyping.

“The real time aspect becomes most valuable in the areas where keeping a state of flow in the creativity space becomes incredibly useful,” Diego Rivas, a Google DeepMind engineer, said during a press briefing. “For example, in the ideation phase, in the storyboarding phase—anytime you’re kind of in need of rapid ideation.”

Google’s existing AI video tool, Veo, already creates high‑definition clips with built‑in synchronized audio from text, image, or video prompts, a feature that has gained traction among filmmakers. Looking ahead, Rivas noted that users could eventually rely on Gemini Omni, a multimodal large language model capable of generating video from images, audio, and text, to transform these prototypes into polished creative assets. He cautioned that such capabilities remain some distance away.

Despite its advances, Genie 4 still has limitations. Details and objects can drift as a simulation extends, and the team has yet to fully guarantee audio fidelity or legibility of in‑world text.

Genie 4 is currently available as a research preview for trusted testers. In the coming weeks, the model will be rolled out to Google AI Ultra subscribers through Project Genie, Google’s platform for creating interactive worlds from text or image prompts, and to early‑access partners of the Gemini API and Google AI Studio.