Demis Hassabis, CEO of Google’s DeepMind, teased a potentially groundbreaking direction for Google’s latest video-generating AI model, Veo 3, hinting it could someday power interactive video game worlds.
On Tuesday evening, Hassabis reacted to a social media comment suggesting Veo 3 should evolve into playable game environments, quipping, “now wouldn’t that be something.” Following that remark, Google’s AI Studio lead product manager Logan Kilpatrick joined in the conversation, playfully responding with cryptic emojis without disclosing further information.
Despite the informal nature of these comments, the prospect of Google venturing into playable virtual environments using video-generation technologies appears plausible. Currently in public preview, Veo 3 can generate highly realistic video content, along with original audio components—including speech and soundtracks. It achieves realistic movements partly by simulating real-world physics. However, Veo 3 still remains predominantly passive, providing static content rather than interactive simulations. Its immediate applications likely lean more towards cinematic game cutscenes, promotional trailers, or narrative prototyping rather than functioning as a fully interactive game engine.
Historically, world models differ significantly from today’s standard video-generation technologies. While video-generation models excel at synthesizing realistic video sequences, world models dynamically represent an interactive environment, allowing AI-powered agents and players to manipulate and respond to changing variables within those simulated worlds. Google itself plans to focus on multimodal models like Gemini 2.5 Pro to replicate certain capabilities of human cognition within simulated environments.
This isn’t Google’s first exploration into more elaborate world-modeling. Last December, DeepMind unveiled Genie 2, a sophisticated AI able to create an essentially unlimited variety of playable virtual worlds. Shortly afterward, the company established an entirely new team dedicated to developing AI capable of simulating the physical world realistically.
Outside Google, interest in fully interactive AI worlds is accelerating. AI innovator Fei-Fei Li, founder of World Labs, recently showcased a system capable of generating 3D interactive scenes from single-image inputs. Microsoft’s ongoing developments, alongside projects such as Scenario, Runway, Pika, and OpenAI’s upcoming Sora model, further underscore growing competition in this emerging sector.
Should Google choose to advance Veo 3 beyond cinematic and passive generation and focus on active, interactive world-model simulations, its vast resources and significant distribution capability could swiftly position it as a formidable competitor. Such a shift would require significant innovation to ensure consistent, controlled, and real-time simulations essential to creating fully playable AI-driven worlds. Given the tech giant’s history and strategic investments, rivals in the space will undoubtedly watch Google’s movements carefully.
