Google DeepMind launches world model Genie 4, achieving synchronized audio-video generation for the first time

Google Launches Its Most Powerful World Model Genie 4: Create Interactive Virtual Worlds with a Single Sentence, Achieving Synchronized Audio-Video Generation for the First Time

English text is being prepared.

English text is being prepared.

English text is being prepared.

English text is being prepared.

Launching Genie 4 is Google's latest move to stay ahead in the fiercely competitive field.

More than a year ago, Google released the previous-generation model Genie 3 as a research preview. Besides a major boost in visual realism and the addition of audio capabilities, Genie 4 supports free-form editing operations, such as adjusting the weather or camera angle in real time, or adding characters and objects. It can maintain a coherent, unchanging scene for up to two minutes — twice as long as the previous model — with better physics and virtual characters that are closer to real people.

Integration between Google's large model Gemini and Genie 4 has also deepened further: Gemini acts as the director, planning steps and validating frames, while Genie 4 generates the visual content. In a research experiment related to the Gemini robotics project, the Gemini 4 Argon model broke tasks down into multiple steps, which Genie 4 then simulated in real time. The experiment demonstrated a viable approach: validating scenarios in a simulated environment before conducting robot tests in the real world.

"In the future, we'll see more and more of these hybrid world models, where Gemini writes code and text, and Genie 4 reads that content in real time and simulates reality at different levels of abstraction," Tejas Kulkarni, Google's director of research, said at a briefing.

Like other world models, its main commercialization scenarios include gaming, interactive learning, and robotics. One envisioned case for interactive learning is a user entering a painting depicting a historical scene and conversing with the figures in the painting about the events recorded in it.

However, "trusted testers" — the first users able to try Google's frontier products — have found a potential new use: film prototyping.

"The value of real-time characteristics is especially prominent in scenarios that require maintaining a coherent creative thread. For example, the ideation stage, the storyboarding stage — all the links that require rapid bursts of creativity," Diego Rivas, an engineer at Google DeepMind, said at a media briefing.

Google has also launched Veo, an AI video generation model that can generate high-definition short films from text, image, or video prompts, with synchronized audio built in; the tool's popularity among film and television creators continues to rise.

Rivas added that users may in the future be able to use Gemini Omni, a multimodal large model, to turn prototype material into formal creative assets; Gemini Omni can generate video based on images, audio, and text. However, he also noted that these capabilities are still some distance from deployment.

Throughout the briefing, Rivas repeatedly emphasized that Genie 4 is still in its early stages and that what the final full product will look like is not yet settled. The team also acknowledged that Genie 4 still has shortcomings: the longer the simulation runs, the more details and objects tend to drift, and the team has not yet found a reliable way to ensure audio accuracy while keeping text in the virtual world clearly legible.

Genie 4 is currently available in research preview to trusted testers. Over the coming weeks, the model will gradually be opened to Google AI Ultra subscribers and the Genie project — an AI tool Google launched that can generate interactive virtual worlds from text or image prompts — and early partners of the Gemini API and Google AI Studio will also gain access.