Google Launches World Model Genie 4, Doubling Simulation Duration to 2 Minutes and Achieving Synchronized Audio-Video Generation for the First Time

Google has launched its next-generation world model, Genie 4, extending the maximum simulation duration to 2 minutes and achieving synchronized audio-video generation for the first time; users can also change the virtual environment, control the camera, and add characters in real time through natural language. Google is exploring its applications in fields such as game design, film and television production, education, and robot training.

On Tuesday, October 27, US Eastern Time, Google announced that Genie 4 is its most realistic and most interactive world model to date. Unlike traditional video generation models that mainly generate video clips based on prompts, Genie 4 can build explorable virtual environments based on text instructions and adjust the scene in real time as users continue to input instructions, allowing users not only to watch AI-generated worlds but also to enter them and interact with the environment.

Google said that Genie 4 is currently available as a research preview to a small number of screened testers, will be opened to Google AI Ultra subscribers in the coming weeks, and will gradually provide access to some developers.

This upgrade comes only months after Google launched Genie 3 and opened the Project Genie experience. Genie 4 further improves the realism, real-time controllability, and multimodal interaction capabilities of generated environments, meaning that this technology is moving from generating explorable virtual scenes toward more complete interactive world simulation.

From generating video to real-time simulation of the world, users can change the environment at any time

World models are fundamentally different from common video generation models. The latter mainly generate a video based on a prompt, while a world model needs to continuously simulate environmental changes and generate subsequent frames in real time according to the user's actions and instructions, so that the virtual world can maintain a certain degree of coherence during interaction.

Genie 4 allows users to dynamically change the simulated environment through natural language. For example, users can ask the scene to switch from sunny to heavy rain, change the camera perspective, or add new characters and objects to the existing environment. The system executes these instructions while the simulation continues running, rather than generating a separate video from scratch each time.

The new model extends the maximum simulation duration to about 2 minutes, roughly twice that of the previous generation, Genie 3. However, a longer simulation time does not mean that the scene always remains completely consistent. Google acknowledges that as the simulation continues, visual details and objects may still drift, and issues such as complex physical interactions, readable text, and maintaining environmental consistency over long periods remain to be solved.

Google DeepMind believes that simulating the real world is an important research direction toward artificial general intelligence (AGI). By building virtual environments that can respond to actions and follow certain physical laws, AI can not only generate content but also learn within them how the environment changes and what outcomes different actions may bring.

Gemini serves as the "brain" of the simulated world, planning and adjusting scenes in real time

Another important upgrade in Genie 4 is the combined generation of audio and visual scenes.

The model can generate ambient sound effects, music, and dialogue that interacts with users in real time for virtual environments. Character responses can be dynamically adjusted according to the content of the conversation and changes in the scene, rather than simply playing pre-recorded sounds. This moves the generated virtual world beyond the visual level and further toward an interactive environment with sound feedback and real-time communication capabilities.

Behind the scenes, Google uses the Gemini model to provide reasoning and planning capabilities for the simulated world. According to Google's introduction, Gemini understands the user's goals, breaks down complex requirements into a series of actions, observes changes in the simulated scene and tracks the progress of events, and then adjusts the execution process according to the user's subsequent instructions.

It is worth noting that, according to reports, the Google team tried to use the latest flagship model Gemini 4 Argon, but the real-time interactive experience of Genie 4 actually runs on the lighter Gemini 3.8 Flash. This shows that real-time world simulation depends not only on the quality of generated images, but also on the model's ability to continuously understand the scene, plan actions, and respond to users with low latency.

The focus of Google's upgrade this time is therefore not only to make the virtual environment look more realistic, but also to enable the environment to change continuously with user input and to establish closer connections among visuals, sound, and interactive behavior.

Exploring deployment scenarios in gaming, film and television, and robotics

Google is exploring practical uses of Genie 4 with screened partners, focusing on film and television production, game design, education, and robot simulation.

In the film and television industry, Genie 4 can be used for visual effects (VFX) previsualization, helping creators quickly build and explore scenes before formal production. Google also mentioned that testers are trying to use verified historical archives to recreate historical scenes, providing new possibilities for history education and immersive content creation.

In game development, designers can build explorable prototype scenes through natural language and adjust environments and characters during interaction. This is expected to shorten the time required for some proof-of-concept and early prototyping work, allowing developers to test different scene designs and gameplay ideas more quickly.

However, Google is still in the exploration stage. DeepMind product manager Diego Rivas said that the company is working with partners to study how this technology can fit into existing workflows and where the boundaries of its capabilities lie. He emphasized that the goal of Genie 4 is to expand the possibilities achievable through existing creative methods, rather than declaring that production processes in the relevant industries have been fully replaced.

Robot training is another potential use. Researchers can use the generated virtual environments to let robots or AI agents practice navigation, object manipulation, and responding to environmental changes in relatively safe simulated scenarios, and then study how to apply the training results to the real world.

But for such simulation to truly serve robot training, the model still needs to further improve the accuracy of physical laws, the long-term consistency of scenes, and the reliability of complex object interactions. A more realistic virtual world does not necessarily mean that the physical behavior within it is more accurate.

The launch of Genie 4 once again brings the potential impact of AI world models on the gaming industry to investors' attention.

In January this year, after Google launched the Project Genie prototype powered by Genie 3, the market began to worry about whether, if AI can automatically generate explorable virtual worlds based on simple text descriptions, some of the scene production and prototype design work required for future game development would be automated, and whether the business models of traditional game engines and related tool companies would be affected. Game-related stocks such as Unity were once sold off.

This time, Genie 4 further enhances real-time environment control capabilities and adds native audio and more realistic character performance, giving such concerns a new technological context. But judging from current progress, world models cannot yet be directly equated with mature game development tools, let alone be used to assert that traditional game development processes will be replaced.

Google acknowledges that Genie 4 is still an experimental technology, and issues such as complex physical interactions, the readability of text generation, and the consistency of scenes over longer periods remain unresolved. Even if the simulation duration reaches two minutes, it does not mean that every detail in the scene can always be accurately sustained.

In addition, there is still a clear gap between generative world models and complete games. Commercial games usually require stable rule systems, repeatable game mechanics, precise character control, and content that developers can maintain and modify over the long term. For real-time generated virtual environments to meet these requirements, technical and productization challenges still need to be overcome.

Therefore, at this stage, Genie 4 is more appropriately regarded as a potential productivity tool in the gaming, film and television, and robotics fields, rather than a mature product already capable of fully replacing them. Whether it can truly change the cost structure and competitive landscape of the gaming industry still depends on subsequent model stability, developer adoption, and commercialization progress.

Phased rollout of access, with all generated content carrying SynthID watermarks

In terms of access arrangements, Google is currently adopting a phased approach. Genie 4 will first be opened to a small number of screened testers, and in the coming weeks it plans to provide access to Google AI Ultra subscribers and open early access to the Gemini API and Google AI Studio to some developers.

Google also emphasized that Genie 4 has strict safety protection mechanisms in place, and all generated audio and video will be embedded with SynthID watermarks. This technology is intended to help identify content generated by Google's AI systems and reduce the risk of generative media being mistaken for real records.

As AI models evolve from generating static images and videos to virtual worlds capable of responding to instructions in real time, safety governance also needs to cover more complex interactive scenarios. Genie 4 is still being launched in the form of a research preview this time, meaning that Google hopes to continue evaluating technical performance, potential risks, and practical application value while expanding the scope of use.

From a longer-term perspective, the focus of competition in world models is shifting from simply improving generated image quality to building interactive systems that can run continuously, respond to actions, and simulate environmental changes. Genie 4's upgrades in simulation duration, sound interaction, and real-time control demonstrate Google's latest progress in this direction, but there is still a way to go before reaching reliable, large-scale commercializable general simulation environments. Risk disclosure and disclaimer

Markets carry risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial conditions, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their particular circumstances. Investment based on this is at the user's own risk.