Could Google's Veo 3 mark the beginning of playable world models?

Demis Hassabis, CEO of DeepMind, Google's artificial intelligence research organization, hinted on Tuesday night that Veo 3, Google’s latest video generation model, could be used in the realm of video games. This came in response to a post on X calling for Google to allow gameplay using videos generated by Veo 3 and asking about the availability of "playable world models." Hassabis commented that "that would be interesting."

The next day, Logan Kilpatrick, product lead for Google AI Studio and Gemini API, also responded with a silent emoji. However, the comments from both executives seem more like a joke than a serious assertion, and a Google spokesperson communicated that there are no updates at this time. Nevertheless, the development of playable world models is not a far-fetched idea for the company.

World models differ from video generation models in that the former simulate the dynamics of a real environment, allowing agents to predict how that environment will evolve in response to their actions, while video models create realistic visual sequences. Google plans to transform its multimodal model, Gemini 2.5 Pro, into a world model that mimics aspects of the human brain.

In December, DeepMind introduced Genie 2, a model capable of generating an "infinite" variety of playable worlds. The following month, it was reported that Google was forming a new team to develop AI models that could simulate reality. Others, like AI pioneer Fei-Fei Li, are also working on world models; Li launched her startup World Labs last year, which has created its own AI system to generate 3D scenes reminiscent of video games from a single image.

Veo 3, which is still in public beta, can generate both video and audio to accompany clips, including everything from speeches to soundtracks. Although Veo 3 creates realistic movements by simulating real-world physics, it is not yet a world model. It could be useful for cinematic storytelling in games, such as cutscenes, trailers, and narrative prototypes.

This model remains a "passive" output tool, and for future versions of Veo to engage in more active, interactive, and predictive simulations, they would need to evolve. However, the real challenge in video game production doesn't solely lie in having stunning graphics; real-time, consistent, and controllable simulation is necessary. Therefore, it might be reasonable for Google to adopt a hybrid approach that combines Veo and Genie in the future if it decides to venture into developing video games or playable worlds.

Google could face competitors like Microsoft, Scenario, Runway, Pika, and eventually OpenAI's video generation model Sora. Given Google's intentions in developing world models and its established ability to outpace rivals, it would be wise for competing companies to keep a close watch on its moves.