Google (GOOGL.US) Releases Next-Generation AI World Model Genie 4 to Generate Interactive Virtual World Game Industry in Real Time or Embrace Changes

The Zhitong Finance App learned that Google's parent company Alphabet (GOOGL.US, GOOG.US) announced on Tuesday the launch of Genie 4, a next-generation artificial intelligence world model, which can generate more realistic interactive virtual environments in real time based on text instructions, and allows users to dynamically modify scenes during exploration. Recent advances in this technology have once again drawn market attention that AI may change the video game development model. Earlier this year, when Google first introduced related technology, it caused many game companies' stock prices to drop sharply.

According to reports, Genie 4 was developed by Google DeepMind, Google's artificial intelligence research division. Users can simply type in text prompts to create a virtual world they can freely explore and control movement through it using the keyboard. Unlike traditional AI video generation tools, Genie 4 can not only generate images, but also adjust the virtual environment in real time according to user actions, making it more interactive.

Additionally, users can change scenes by entering new text commands while the virtual world is running, such as adjusting weather conditions or adding new characters, without having to regenerate the entire environment. Compared to the previous Genie 3, Genie 4 has improved both simulation time and multi-modal generation capabilities.

According to Google, the interactive virtual environment generated by Genie 4 can run for up to two minutes, which is about twice that of Genie 3. At the same time, Genie 4 is also the first model in the series that can simultaneously generate images and sounds, supporting ambient sound effects, music, and character conversations that respond to user actions.

Currently, Google is opening up Genie 4 to a small number of trusted testers in the form of a research preview. In the coming weeks, the company plans to further make the model available to users subscribed to the highest-level Google AI Ultra paid plan, as well as to some developers developing applications using Google AI Studio and Gemini APIs.

A “world model” (World Model) refers to an artificial intelligence system capable of simulating an environment and its dynamic changes. Unlike traditional text-generated video models that mainly output pre-generated video content, the world model can continuously generate and adjust virtual scenes according to the user's real-time operations. This technology is expected to be used in fields such as video game development, film and television production, virtual environment design, and robot training.

When Google first launched the Genie 3-powered Project Genie prototype in January of this year, it raised Wall Street's concerns about the future of the video game industry. As investors feared that AI-generated virtual world technology might disrupt traditional game development processes, the stock prices of many game-related companies fell sharply at the time.

As Genie 4 further enhances real-time interaction and multi-modal generation capabilities, the market's concern that AI technology may change the way game content is produced also heats up.

In the field of artificial intelligence, Google is facing intense challenges from competitors such as OpenAI and Anthropic. However, with the recent launch of the Gemini 4 Argon model, the market believes that Google is closing the technology gap with its competitors.

At the same time, the world model and multi-modal AI have long been viewed as an important technological advantage for Google. Multimodal technology enables artificial intelligence to simultaneously process different types of information such as text, images, audio, and video, and world models further apply these capabilities to simulating dynamic environments.

Notably, Genie 4 also uses the Gemini model to enhance the coherence and interactive performance of the virtual world. Google explained in its official blog that when Genie 4 generates a virtual scene, Gemini will continue to observe the newly generated images, understand the goals the user wants to accomplish, disassemble them into a series of specific steps, and then check whether the relevant operations have been achieved.

This mechanism helps the model understand user intent more accurately and maintain a degree of logical continuity in a dynamically generated virtual environment. Google said that the R&D team tried using the Gemini 4 Argon model, but currently Genie 4's real-time interactive experience is mainly supported by the Gemini 3.8 Flash model, which has higher computational efficiency.

In terms of practical applications, Google has invited testers, including game developers and filmmakers, to experience Genie 4, explore how to quickly transform initial ideas into interactive virtual prototypes, and use AI to generate scene sketches to aid content creation.

Furthermore, Google is also studying whether the world model can be used for robot training. By simulating different environments and physical scenarios, related technologies may help robots learn and test behavior in the virtual world in the future.

However, Google emphasized that it is currently uncertain whether these potential apps will be widely adopted. In response to Genie 4's possible impact on the video game industry, Diego Rivas, product manager at Google DeepMind, said in a press briefing that the company is working with partners to study how this technology can be integrated into existing workflows and the boundaries of its actual capabilities.

Rivas pointed out that Google hopes to further expand the functions that can be achieved with existing tools through world model technology, rather than simply replacing traditional content creation methods. Although Genie 4 has made significant progress, Google also acknowledged that the model is still experimental and is still a long way from becoming a full-fledged commercial product.

Currently, as the virtual scene runs longer, the visual details and objects generated by the model may still change, affecting the consistency of the environment.

Google DeepMind research engineer Hang Qi said that even if Genie 4 can support simulations for up to two minutes, this does not mean that the entire scene remains completely consistent during this period. Furthermore, aspects such as complex physical interaction, clear and readable text generation, and maintaining the stability of the virtual world over a long period of time still need further improvements.

Tejas Kulkarni, head of research at DeepMind, points out that even just keeping the virtual world completely consistent within two minutes requires models to truly understand the laws of physics in the real world.

Rivas also said that the two-minute simulation time is still clearly limited, and Genie 4 is currently more in the technical research and exploration stage than in the final product development stage.

In terms of the identification and security of AI-generated content, Google said that all audio and video generated by Genie 4 will be embedded with an invisible SynthID digital watermark to help identify AI-generated content.

Overall, the launch of Genie 4 marks a new advance for Google in the field of real-time interactive AI world models. Although it is currently still subject to technical limitations such as simulation time, physical consistency, and generation quality, its demonstrated capabilities have drawn the market's attention to the future development direction of industries such as game development, film and television production, and robot training.

As Google gradually expands the testing scope of Genie 4, the market will pay more attention to whether world model technology can evolve from an experimental tool to a product with actual commercial value, and whether it will have a more profound impact on traditional game development and digital content production models.