Google DeepMind on Tuesday released Genie 4, its newest world model, in a research preview limited to what the company calls “trusted testers.” The model generates interactive environments that look more photorealistic than earlier versions, and users can type prompts mid-simulation to change the scene.
Genie 4 also produces real-time audio that matches the simulated world, including music, ambient noise and interactive dialogue. Google says the model can now keep a scene running for about two minutes, up from one minute for its predecessor, Genie 3. Users can make free-form edits while a simulation runs, such as changing the weather or camera angle or adding new people and objects.
The model also works more closely with Gemini, Google’s AI model. In Genie 4, Gemini acts as a planner that breaks tasks into steps and checks frames while Genie creates the visuals. In one research experiment tied to Gemini Robotics, that pairing broke robot tasks into steps that Genie 4 simulated in real time. The experiment points to a way robot scenarios could be tested in simulation before they run in the physical world.
“What we’ll increasingly see in the future are these hybrid world models, where you have a combination of Gemini writing code and text, and Genie ingesting that in real time and simulating reality at different levels of abstraction,” Tejas Kulkarni, research director at Google, said during a briefing.
Google is pitching Genie 4 for gaming, interactive learning and robotics. One speculative learning example is stepping into a painting of a historical scene and chatting with the people in it about the event. Filmmakers, though, appear to be the first group finding a practical use.
Diego Rivas, a Google DeepMind engineer, said the real-time aspect is most useful when creators need to keep their momentum. “The real time aspect becomes most valuable in the areas where keeping a state of flow in the creativity space becomes incredibly useful,” Rivas said during a press briefing. “For example, in the ideation phase, in the storyboarding phase — anytime you’re kind of in need of rapid ideation.”
Google already sells Veo, its AI video-generation model that creates high-definition clips with synchronized audio from text, image or video prompts. Rivas said users could eventually rely on Gemini Omni, a multimodal large language model that can generate video from images, audio and text, to turn Genie prototypes into finished creative assets. He said those capabilities are still a ways off.
Rivas stressed that the product is early and that the final end-to-end experience is not yet defined. The team also acknowledged gaps. Details and objects can drift the longer a simulation runs, and the company has not fully worked out how to make sure audio is accurate or that text inside a world is legible.
The release follows a busy stretch for the sector. Runway recently released its second world model, GWM Worlds 2, which responds dynamically to user inputs and also generates audio. The category is also consolidating. AMD is acquiring World Labs, the company led by Fei-Fei Li. Amazon is in talks to buy Decart, and Rivian recently acquired Third Dimension to bring simulation capabilities in-house for self-driving testing.
Genie 4 is available now only to trusted testers. Google says it is coming soon to Google AI Ultra subscribers, and the company has opened a waitlist for people who want to try it in Project Genie.
Este artículo fue escrito con la asistencia de IA.
News Factory APP - noticias agénticas para impulsar tu SEO y AEO.