Google’s next Genie AI model generates more realistic experiences
Google has unveiled Genie 4, the newest iteration of its world model designed to generate interactive, AI-driven experiences. The update builds on Genie 3 with four core improvements: enhanced realism, real-time editing via natural language prompts, native audio generation, and a new underlying engine powered by Gemini 3.8 Flash. The maximum duration of generated experiences has also doubled, from one minute to roughly two minutes.
The real-time prompt editing is one of the more practical additions. Users can now adjust an experience as it plays out by describing what they want in plain text, rather than needing to regenerate content from scratch. This kind of interactivity moves the model closer to a tool for active creative work rather than a one-shot generation system.
Google is positioning Genie 4 for a range of professional use cases, including pre-visualization for film productions, early-stage game prototyping, and the creation of training environments for robotics. These applications share a common need - quickly producing plausible, interactive scenarios without the cost and time of building them by hand. The audio generation capability adds another layer to that utility, making outputs more self-contained and closer to a finished reference material.
The choice to integrate Gemini 3.8 Flash as the backbone reflects a broader trend of Google tying its world models more tightly to its wider Gemini ecosystem. Genie has been an ongoing research and product effort at Google DeepMind, focused on learning interactive world models from video data. With Genie 4, the project appears to be moving from a primarily research-oriented footing toward tools with clearer, near-term practical applications across creative and technical industries.
