Google DeepMind has launched its latest AI world model, Genie 4, in a research preview. The model aims to generate more realistic and interactive experiences, significantly advancing capabilities over its predecessor, Genie 3, according to reports by The Verge and TechCrunch.
Genie 4 introduces several enhancements, including the ability to create more photorealistic worlds and generate audio that matches the simulated environment, such as music, ambient sounds, and interactive dialogue, as noted by The Verge and TechCrunch. Users can now update an experience in real time using natural language prompts, allowing for freeform edits like changing weather, camera angles, or adding objects and people, TechCrunch reported.
The new model can maintain a consistent scene for up to two minutes, doubling the one-minute limit of the previous Genie 3 model, both The Verge and TechCrunch indicated. TechCrunch added that Genie 4 also features better physics and more human-like characters. Genie 4 leverages Google's Gemini AI model, with TechCrunch reporting that Gemini acts as a director, planning steps and checking frames as Genie creates visuals. The Verge added that Genie 4 uses Gemini 3.8 Flash to assist with simulations.
Google pitches Genie 4 as a tool for various professional applications. These include pre-visualization for filmmakers, prototyping game ideas, and creating training scenarios for robots, both The Verge and TechCrunch reported. TechCrunch highlighted potential for interactive learning, such as entering a historical painting, and for rapid ideation and storyboarding in creative fields. Google's research director Tejas Kulkarni, as reported by TechCrunch, anticipates a future with hybrid world models combining Gemini's text and code generation with Genie's real-time simulation.
Despite these advancements, Google acknowledges that Genie 4 is in its early stages. Details and object permanence can shift over time as simulations run longer, a point noted by both The Verge and TechCrunch. TechCrunch further reported that the development team has not yet fully addressed challenges in ensuring audio accuracy or the legibility of in-world text. Diego Rivas, a Google DeepMind engineer, emphasized the value of real-time capabilities for maintaining creative flow, particularly during ideation and storyboarding phases, according to TechCrunch. However, the ultimate end-to-end product experience remains unclear.
Genie 4 is currently available as a research preview for trusted testers, according to The Verge and TechCrunch. Google has opened a waitlist for those wishing to try it in Project Genie, The Verge noted. The model is expected to roll out to Google AI Ultra subscribers in the coming weeks and will also be available to early access partners of the Gemini API and Google AI Studio, as reported by TechCrunch.
The release of Genie 4 positions Google to stay competitive in the evolving world model industry, TechCrunch noted. It highlighted that AI video startup Runway recently launched its GWM Worlds 2, a playable model with audio generation, and mentioned consolidation in the sector with acquisitions like World Labs by AMD and Third Dimension by Rivian.
