Demis Hassabis, CEO of Google’s AI research institute Deepmind, appears to have hinted that Google’s latest video generation model, Veo 3, could potentially be used for video games on Tuesday evening.
When I asked Google, “Let me play a video game with VEO 3 video,” Hassavis replied, “Wan a playable world model?”
On Wednesday morning, Logan Kilpatrick, the lead product of Google’s AI Studio and Gemini API, rang out with the reply, “🤐🤐🤐🤐.”
Both posts from Google executives are nothing more than playful suggestions, and a Google spokesperson told TechCrunch there is nothing to share right now. But building playable world models is not outside the realm of potential for tech giants.
World models differ from video generation models. The former simulates the dynamics of real-world environments. This allows agents to predict how the world will evolve according to their actions. The Videogen model synthesizes realistic video sequences.
Google plans to turn the Gemini 2.5 Pro, a multimodal foundation model, into a world model that simulates aspects of the human brain. In December, DeepMind announced the Genie 2. This is a model that can generate a variety of “infinite” playable worlds. The following month, Google reported that it was forming a new team to tackle AI models that could simulate the real world.
Others are working on building a world model. Most notably, the AI Pioneer Fei-Fei Li. Li came out from Stealth at World Labs last year. WorldLabs is a startup that has built its own AI system that generates video game-like 3D scenes from a single image.
Still available, VEO 3 allows you to create video and audio to match the clip, from speeches to soundtracks. VEO 3 creates realistic movement by simulating real physics, but that is not yet a world model. Instead, it can be used for film storytelling in games, such as cutscenes, trailers, and story prototyping
The model is still a “passive output” generation model, and IT (or future VEO generations) should move to a more active, interactive and predictive simulator.
But the real challenge in video game production is not just merely impressive visuals. A real-time, consistent, controllable simulation. That’s why it might make sense to see Google take a hybrid approach that leverages Veo and Genie in the future when pursuing video games and playable world development.
Google was going to compete with Microsoft, Scenario, Runway, Pika and, ultimately, Openai’s video generation model Sora.
Given the movements Google plans in the world model space and its reputation for using its deep pockets and distribution muscles to roll its rivals with steam, competitors in this space would be wise to pay attention.
Source link
