Google DeepMind has introduced Genie 4, an artificial intelligence model that creates interactive virtual environments users can modify while exploring, adding more realistic visuals, responsive audio, and greater control over simulated scenes.
Announced on October 27, 2026, the model is initially available to selected testers through a research preview. Google plans to extend access to Google AI Ultra subscribers through Project Genie in the coming weeks, alongside early-access partners using the Gemini API and Google AI Studio.
Unlike conventional AI video tools that generate clips from prompts, Genie 4 allows users to intervene during a simulation. They can issue text instructions to change elements such as the weather, camera angle, characters, or objects while the environment continues to evolve.
Longer simulations with sound and improved realism
Genie 4 builds on Google’s earlier Genie 3 model, extending the duration of consistent simulations from roughly one minute to up to two minutes.
The latest version also introduces real-time audio designed to match events within the generated environment. This includes background sounds, music, and interactive dialogue, making the virtual worlds more immersive.
Google says the model has improved physics and more lifelike characters. However, some limitations remain: objects and scene details can become inconsistent as simulations continue, while audio accuracy and the readability of text inside generated environments are still unresolved challenges.
Gemini helps direct the virtual environment
Google is also bringing its Gemini AI models closer to the simulation process. Gemini can plan actions and inspect frames, while Genie generates the environment and responds to instructions.
In a research experiment connected to Gemini Robotics, a Gemini 4 Argon model divided tasks into steps that Genie 4 then simulated in real time. The approach points to a possible way to test robotic actions in virtual environments before attempting them in the physical world.
Google research director Tejas Kulkarni described a future in which Gemini handles tasks such as generating text and code, while Genie simulates the resulting scenarios at different levels of detail.
Such a combination could help developers explore situations that would be expensive, slow or difficult to reproduce in the real world.
Filmmakers and game developers could benefit
Google sees potential applications for world models in gaming, interactive education and robotics. The technology could let users explore virtual environments rather than simply watch pre-rendered videos.
Filmmakers may also find the model useful during early creative development. Google DeepMind engineer Diego Rivas said real-time interaction could speed up brainstorming and storyboarding by allowing creators to experiment with scenes and ideas as they emerge.
Google already offers Veo, its AI video-generation model, which produces high-definition clips with synchronised audio from text, image or video prompts. The company suggested that tools such as Gemini Omni could eventually help turn concepts developed in Genie into finished creative assets, although that workflow is not yet ready.
Competition in AI-generated worlds intensifies
Genie 4 enters a growing market for AI systems that generate environments users can actively influence.
AI video company Runway introduced GWM Worlds 2 in September, a world model designed to generate interactive simulations with continuous video and audio. The wider sector is also attracting interest from major technology and automotive companies seeking simulation capabilities for applications such as robotics and autonomous driving.
These developments are increasing competition to build AI systems that can model environments, respond to user actions and support more than passive video generation.
Access remains limited as Google tests the technology
For now, Genie 4 is not broadly available. Google is providing access to trusted testers before expanding availability through Project Genie and selected developer platforms.
The company has not yet settled on the model’s final product experience, and its acknowledged technical limitations suggest further development is needed before it can reliably support every proposed use case.
For creators and developers, the release offers an early look at AI environments that can be explored and reshaped in real time. Whether Genie 4 becomes a practical tool for production, education or robotics will depend on how well Google addresses those remaining limitations.
Source: TechCrunch