Google launches Genie 4: Now create virtual worlds using AI—learn how to chan.ge weather and characters in real time.

Google DeepMind has officially launched its new world AI model, Genie 4. This next-generation AI system is capable of creating real-time, interactive 3D worlds, complete with native audio and highly realistic visuals.

According to the company, users can not only generate dynamic environments but also navigate through them in real time. The model calculates the underlying laws of physics, character behaviors, and soundscapes. Sundar Pichai, CEO of Alphabet and Google, shared the news, stating that Genie 4 is their most realistic and interactive world model to date.

What is Google Genie 4, and how does it work?
Genie 4 is an AI world model. It can create digital environments where users can engage in activities and observe the immediate effects of their actions. While standard photo or video generators typically produce static, pre-defined content, a world model aims to create an environment that evolves dynamically based on user input and actions.

For instance, a user could generate a seaside scene and navigate through it. They could then use text prompts to change the weather, shift the camera angle, or introduce new characters. The model then adapts the scene in response to the user's requests.

Google explains that Genie 4 can simultaneously simulate various elements—such as how a world looks and sounds, as well as how the characters or individuals within it move and react.

Modify virtual worlds in real-time.
One of the standout features of Genie 4, according to the company, is the ability to modify scenes in real-time. Users can control navigation and camera movement via their keyboard. Furthermore, the simulation can be altered dynamically through the use of text prompts. With this, users can issue commands—such as changing the weather, adjusting camera angles, or adding new characters and objects. The model interprets these instructions in the context of the surrounding scene and evolves the virtual world accordingly.

This feature could prove highly useful for game developers and creative professionals. For instance, a game development team could transform an initial concept into a virtual world, allowing them to evaluate its visual aesthetic, environment, and interactive elements. This facilitates the creation of early prototypes and the exploration of various ideas.

Focus on Physics and Character Movement
Google states that 'Genie 4' features more realistic-looking scenes and improved movement compared to its predecessors. It offers enhancements in areas such as aerodynamics, collision dynamics, and environmental interactions.

Attention has also been paid to how characters move, react, and are influenced by their surroundings. The goal is not merely to create a visually realistic representation of the world, but to build a simulation where the consequences of actions are clearly reflected.
However, Google has clarified that 'Genie 4' is still in the early research phase. Over extended simulation periods or through repeated modifications, the appearance and positioning of objects may shift. There is also a continued need for improvement in complex physical interactions and spatial accuracy.

Enhanced Native Audio Capabilities
'Genie 4' places a strong emphasis on native audio generation. This means the model can produce sounds that correspond to the virtual world—ranging from ambient environmental noises and background music to interactive dialogue between characters.

For example, one might hear the sounds of birds in a virtual forest or the chatter of people in a marketplace scene. As the scene shifts based on user commands, the corresponding audio is generated dynamically to match. Google counts this among the key capabilities of Genie 4. However, according to the company, audio accuracy is still an area in need of improvement; therefore, no guarantee sound and action will be perfectly synchronized in every scene.

Creating multi-step events with the help of Gemini
Genie 4 has been integrated with the reasoning capabilities of Gemini models. According to the company, 'Gemini Omni' helps bridge the gap between the simulation and the appearance and behaviors of the real world, while other Gemini models play a role in maintaining the sequence of events and the continuity of the simulation.

In this setup, 'Gemini 1.5 Flash' (referred to here as Gemini 3.8 Flash) acts as a guiding agent. It can observe a scene, break down a larger objective into smaller, manageable steps, and evaluate upcoming frames. If a user changes their instructions mid-process, or if the scene itself shifts, the agent can adapt its plan for the next steps accordingly.

Google has also highlighted experiments involving the combined use of Gemini and Genie 4. In this context, the potential is being explored to simulate robotics-related tasks within a virtual environment by breaking them down into distinct, sequential steps.

Such simulations can be instrumental in testing the behavior of an agent or robot—and how it might respond to various scenarios—before it is deployed in the real world.

Where can Genie 4 be used?
Game Development: For creating initial prototypes of game worlds, characters, and interactions.

Film and Creative Work: For exploring various iterations of a scene or environment.

Robotics: For testing the behavior of robots and AI agents in a virtual environment before deploying them in the real world.

Training Simulations: For evaluating the capabilities and responses of AI agents under various conditions.

Environmental Events: For understanding the potential impact of events like floods or forest fires within a virtual environment.

When and to whom will Genie 4 be available?
Google DeepMind has currently released Genie 4 as a research preview to a limited number of trusted testers. The company intends to gather feedback from these testers first.

Following this, there are plans to make it available to Google AI Ultra subscribers (via Project Genie) and to select early-access partners (through the Gemini API and Google AI Studio). A date for wider availability to the general public has not yet been announced.

How will AI-generated content be identified?
According to Google, videos and audio created using Genie 4 will include a 'SynthID' digital watermark. This watermark is not typically perceptible to the human eye or ear, but it can be used to help identify content generated by AI.

Disclaimer: This content has been sourced and edited from Amar Ujala. While we modified it for clarity and presentation, the original content belongs to its respective authors and website. We do not claim ownership of the content.