Google DeepMind has unveiled Genie 4, the latest iteration of its world model technology, launching it as a research preview this week. The new model represents a significant leap forward in generating interactive, photorealistic environments, and introduces several capabilities that its predecessor lacked - most notably, real-time audio generation and the ability to modify a simulation on the fly through mid-session text prompts.

The audio component is particularly noteworthy. Genie 4 produces sound that corresponds directly to whatever is happening within the simulated world, spanning background music, ambient environmental sounds, and interactive dialogue. This level of sensory integration marks a new stage in what world models can deliver to users and developers alike.

A Growing and Consolidating Industry

The timing of this release places it alongside a broader wave of activity in the world model space. Runway, the AI video startup, launched its own second-generation world model - GWM Worlds 2 - just a couple of months ago. That model similarly responds to user inputs in real time and incorporates audio generation, signaling that these features are quickly becoming baseline expectations in the field.

Meanwhile, the world model industry is showing signs of consolidation. World Labs, the company co-founded by AI pioneer Fei-Fei Li, is in the process of being acquired by AMD. Amazon is reportedly in discussions to purchase Decart. And electric vehicle manufacturer Rivian has acquired Third Dimension, a move aimed at bringing simulation infrastructure in-house for self-driving vehicle testing. Against this backdrop, Genie 4 represents Google's effort to maintain a leading position in an increasingly competitive sector.

What Genie 4 Can Do

Google's previous release, Genie 3, arrived as a research preview roughly a year ago. Genie 4 builds on that foundation in several meaningful ways. Beyond the improved visual fidelity and added audio, the new model supports freeform, real-time edits to a running simulation. Users can alter weather conditions, shift the camera angle, or introduce new characters and objects without breaking the scene. The model can now sustain a coherent simulation for up to two minutes - twice the duration that Genie 3 was capable of - and delivers more convincing physics behavior along with more lifelike human character rendering.

One of the more architecturally significant changes is the tighter integration of Gemini, Google's large-scale AI model, into the Genie 4 pipeline. In this setup, Gemini functions as a director of sorts - planning sequences and evaluating individual frames - while Genie handles the actual visual output. This division of responsibilities is designed to improve coherence and consistency across longer simulations.

A research experiment connected to Google's Gemini Robotics initiative demonstrated one potential application of this architecture. A Gemini 4 Argon model was used to break down a physical task into discrete steps, which Genie 4 then simulated in real time. The experiment illustrates a potential workflow where robotic scenarios are tested and refined in simulation before being executed in the physical world - a capability with obvious appeal to robotics developers and researchers.

"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."

Commercial Applications and Creative Use Cases

Like other world models currently in development, Genie 4 is being positioned around several commercial verticals. Gaming, interactive education, and robotics simulation are the primary categories that Google has highlighted. One illustrative example offered for the education space involves uploading an image of a historical painting and then interacting in real time with the figures depicted in the scene - effectively turning a static artwork into a conversational learning experience.

However, a different use case has been emerging organically from the group of "trusted testers" - the early-access users who receive priority access to Google's most experimental products. These users have been gravitating toward Genie 4 as a tool for prototyping film concepts. The ability to rapidly visualize and iterate on scenes in real time appears to be particularly well-suited to the early ideation and storyboarding stages of creative production.

Diego Rivas, an engineer at Google DeepMind, spoke to this potential during a press briefing. "The real time aspect becomes most valuable in the areas where keeping a state of flow in the creativity space becomes incredibly useful," he said. "For example, in the ideation phase, in the storyboarding phase - anytime you're kind of in need of rapid ideation."

It is worth noting that Google already operates Veo, a separate AI video generation model capable of producing high-definition video clips with synchronized audio from text, image, or video inputs. Veo has gained traction among filmmakers and content creators, and it occupies a somewhat different position in Google's AI product lineup - focused on polished output rather than real-time interactive simulation.

Rivas also suggested that future users could potentially use Gemini Omni - a multimodal large language model capable of generating video from combinations of image, audio, and text inputs - to convert Genie 4 prototypes into finished creative assets. He was careful to note, however, that this kind of end-to-end workflow remains some distance away from practical implementation.

Limitations and What Comes Next

Throughout the briefing, the Google DeepMind team was candid about the current limitations of Genie 4. Rivas repeatedly emphasized that the product is still in its early stages, and that the team has not yet determined what the full end-user product experience will look like. Known issues include a tendency for objects and fine details to drift or degrade as simulations extend in duration. Audio accuracy also remains inconsistent, and any text rendered within a simulated world is not reliably legible. These are acknowledged gaps that the team is working to address.

In terms of availability, Genie 4 is currently accessible to trusted testers through its research preview. Over the coming weeks, Google plans to expand access to subscribers of Google AI Ultra through Project Genie - the company's dedicated platform for generating interactive worlds from text or image inputs. Early access partners working through the Gemini API and Google AI Studio will also be included in the rollout as it progresses.