Google DeepMind has unveiled Genie 4, its latest artificial intelligence (AI) world model, capable of generating realistic, interactive environments that users can explore and modify using text prompts. The new model introduces native audio, improved physics, human-like characters, and longer simulations.

Announced on October 27, 2026, Genie 4 builds on the capabilities of Genie 3, taking AI-generated environments beyond static visuals and videos by allowing users to influence what happens inside a simulated world in real time.

However, the model is currently available only to a limited group of trusted testers, with wider access planned for Google AI Ultra subscribers and developers.

TL;DR

 

• Google DeepMind introduced Genie 4, an AI model that generates realistic, interactive worlds from prompts.
• Users can change environments in real time, including weather, camera angles, objects, and characters.
• Genie 4 adds native audio, interactive dialogue, improved physics, and simulations lasting around two minutes.
• The model is currently in research preview, with broader access coming soon.

Google Genie 4 Brings Real-Time Control To AI-Generated Worlds

According to Google DeepMind, Genie 4 is a new generation of world simulation technology designed to generate interactive environments that respond to user instructions.

Source

Unlike conventional AI video generation models that primarily produce predetermined sequences, Genie 4 allows users to explore and modify environments while the simulation is running.

Users can provide natural-language prompts to change the weather, introduce new characters or objects, adjust camera perspectives, or alter events taking place inside the generated world.

The system uses Gemini models to interpret instructions in the context of the existing environment, helping maintain continuity as the simulation evolves.

Google says Genie 4 also improves the simulation of real-world physics, including collisions and aerodynamics, while generating characters that move naturally and react to their surroundings.

Another major addition is native audio generation.

As explained by Google, Genie 4 can generate environmental sounds, music, and interactive dialogue alongside visuals.

This means a virtual beach can feature crashing waves, while simulated characters can respond to conversations and events unfolding around them.

In its research announcement, Google DeepMind researchers Raia Hadsell, Diego Rivas, and Hang Qi described the technology's broader objective.

"Understanding and simulating the world is a crucial aspect of building AGI," they wrote, referring to artificial general intelligence.

Genie 4 Uses Gemini 3.8 Flash And Doubles Simulation Duration

Google has integrated its Gemini family of AI models into Genie 4 to improve how simulated environments behave and respond to user instructions.

Gemini Omni helps ground simulations in real-world appearance and behaviour, while Gemini 3.8 Flash operates as a guiding agent that monitors the environment, breaks down complex objectives, and tracks whether individual actions have been completed.

This allows the system to adapt its responses as users change instructions or events unfold inside an environment.

The company has also experimented with integrating Genie 4 and Gemini 4 Argon for robotics-related simulations, demonstrating how AI systems could rehearse multistep tasks inside virtual environments.

Meanwhile, The Verge reported that Genie 4 can generate interactive experiences lasting approximately two minutes, compared with around one minute for Genie 3.

This effectively doubles the duration of its predecessor's simulations.

However, longer simulation time does not necessarily guarantee consistent results. Google acknowledged that objects, visual details, and physical interactions can still become inconsistent as environments evolve.

Google Targets Gaming, Filmmaking, Robotics And Other Industries

Beyond creating interactive virtual environments, Google believes Genie 4 could support several commercial and research applications.

For game developers, the technology could make it possible to turn an early concept into an explorable prototype using natural-language prompts before writing code for a game engine.

Filmmakers could use the model for pre-visualization, experimenting with scene composition, lighting, camera movements, character actions, and dialogue before beginning production.

Google also highlighted potential applications in journalism and education, where historical photographs and events could be transformed into explorable simulations.

In robotics, researchers could use generated environments to evaluate how AI agents respond to different situations before deploying them in physical environments.

Google DeepMind is also exploring how these simulations could help model environmental events, including floods and wildfires.

However, these applications remain experimental, and Google has not announced a general commercial release for the model.

When Will Google Genie 4 Be Available?

Google has begun rolling out Genie 4 as a research preview for selected trusted testers.

The company plans to bring the model to Google AI Ultra subscribers through Project Genie, alongside early-access partners using the Gemini API and Google AI Studio.

Google has also opened a waitlist for users interested in accessing Genie 4 through Project Genie.

However, it has not announced a specific date for broader availability.

The model also continues to face technical limitations, including inconsistencies in object permanence, spatial accuracy, complex physical interactions, readable text, and real-time audio accuracy.

Google says its generated video and audio will include SynthID, the company's digital watermarking technology intended to help identify AI-generated content.

For now, Genie 4 represents Google's latest effort to develop AI systems capable of simulating interactive environments rather than simply generating images or videos.

Whether the technology can deliver consistently reliable experiences for gaming, creative production, and robotics will depend on how effectively Google addresses its remaining limitations.