Introducing Google’s Genie 4 World Model: Advanced Audio Enhancements, Improved Realism, and Real-Time Navigation Capabilities
On Tuesday, Google DeepMind introduced its latest world model, Genie 4, during a research preview event. This state-of-the-art model aims to create the most lifelike and immersive environments to date, enabling users to input prompts in real-time during simulations to modify scenes. In addition, it provides real-time audio, enhancing the virtual experience with music, ambient sounds, and interactive dialogue.
The announcement of Genie 4 follows closely behind AI video startup Runway’s launch of GWM Worlds 2, a versatile model that adapts fluidly to user interactions and includes audio generation capabilities. At the same time, the world model sector is experiencing consolidation, with AMD acquiring Fei-Fei Li’s World Labs, Amazon negotiating to purchase Decart, and Rivian taking over Third Dimension to bolster its simulation capabilities for autonomous driving tests.
The debut of Genie 4 reinforces Google’s dedication to preserving its competitive position in this rapidly evolving industry.
Genie 4 is an advancement over its predecessor, Genie 3, which was launched as a research preview just over a year ago. In contrast to Genie 3, Genie 4 offers significantly enhanced photorealism and audio functionalities, along with the capacity to make real-time freeform adjustments such as changing weather patterns, modifying camera angles, or incorporating new characters and objects. It can maintain a coherent scene for up to two minutes—twice the previous duration—while improving physics and character realism.
Google’s Gemini AI model is now closely integrated with Genie 4, functioning as the orchestrator that manages sequences and tracks frames as Genie produces visuals. In a related research initiative with Gemini Robotics, the Gemini 4 Argon model segmented tasks into manageable components, which Genie 4 simulated in real-time, showcasing an effective method for testing robotic scenarios in a controlled environment prior to real-world implementation.
“In the future, we might see the rise of hybrid world models, wherein Gemini merges coding and textual inputs while Genie simulates reality in real-time across various abstraction levels,” observed Tejas Kulkarni, Google’s research director, during a briefing.
Like other world models, potential commercial applications encompass gaming, interactive education, and robotics. A novel use case in educational settings could involve immersing oneself in a historical painting and interacting with its figures to uncover the events portrayed.
However, “trusted testers”—those granted early access to Google’s advanced features—are suggesting an additional innovative application: film prototyping.
“The real-time capability is especially beneficial in creative fields where achieving a state of flow is essential,” stated Diego Rivas, a Google DeepMind engineer, at a press conference. “This is particularly relevant during ideation and storyboarding, or whenever quick brainstorming is necessary.”
Google also showcased Veo, its AI video generation model that creates high-definition clips with synced audio from text, image, or video prompts—a tool that is becoming increasingly popular among filmmakers.
Rivas noted that users might soon rely on Gemini Omni, a multimodal large language model that can generate video from images, audio, and text, to turn those prototypes into creative assets. Nonetheless, he cautioned that these capabilities may take some time to fully emerge.
Throughout the briefing, Rivas stressed that Genie 4 is still in its early phases, and the overall product experience has yet to be fully refined. The team acknowledged that Genie 4 encounters ongoing challenges: as simulations advance, details and objects might misalign, and it battles with maintaining audio fidelity and clarity of in-world text.
Currently, Genie 4 is available as a research preview for trusted testers. In the next few weeks, it will be rolled out to Google AI Ultra subscribers involved in Project Genie, Google’s initiative to create interactive environments from text or image prompts, as well as early access partners of the Gemini API and Google AI Studio.
Purchasing through links in our articles may earn us a small commission. This does not affect our editorial independence.


