Google Veo 3: Playable World Models - A New Era?

Google's Veo 3 and the Potential for Playable Worlds

Demis Hassabis, the CEO leading Google’s AI research division, DeepMind, hinted on Tuesday evening that Veo 3, Google’s newest video generation model, might have applications within the realm of video games.

Responding to User Interest

The discussion began with a request on X, formerly Twitter, urging Google to allow users to create playable video games from their Veo 3 generated videos. The post inquired about the development of “playable world models.”

Hassabis’s response, simply stating “now wouldn’t that be something,” sparked considerable interest.

Further Indication from Google AI Studio

Adding to the speculation, Logan Kilpatrick, the lead product manager for Google’s AI Studio and Gemini API, followed up with a series of zipped-mouth emojis. This further fueled anticipation regarding the possibility.

Official Stance and Current Limitations

Despite the suggestive posts, a Google spokesperson clarified to TechCrunch that the company currently has no announcements to make. However, the creation of playable world models remains a feasible goal for the technology company.

Understanding World Models vs. Video Generation

It’s important to distinguish between world models and video-generation models. World models focus on simulating the dynamics of a real-world environment, enabling agents to anticipate how the world will react to their actions.

Conversely, video-generation models, like Veo 3, are designed to synthesize realistic video sequences.

Google's Broader AI Strategy

Google intends to evolve its multimodal foundation model, Gemini 2.5 Pro, into a world model capable of simulating aspects of human brain function.

In December, DeepMind introduced Genie 2, a model designed to generate a diverse range of playable worlds.

Furthermore, reports surfaced in January indicating Google was establishing a dedicated team to focus on AI models capable of simulating real-world scenarios.

Competition in the World Model Space

Other organizations are also actively developing world models, notably AI pioneer Fei-Fei Li and her startup, World Labs. World Labs has created an AI system that generates 3D scenes resembling video games from a single image.

Veo 3's Capabilities and Future Potential

Currently in public preview, Veo 3 can generate both video and accompanying audio, including speech and soundtracks.

While Veo 3 simulates real-world physics to create realistic movements, it isn’t yet a fully realized world model.

Its current functionality lends itself well to cinematic elements within games, such as cutscenes, trailers, and narrative prototyping.

The Shift to Interactive Simulation

The model remains a “passive output” generator. To become truly applicable to game development, it would need to evolve into a simulator that is more active, interactive, and capable of prediction.

A Hybrid Approach

The primary challenge in video game production lies in achieving real-time, consistent, and controllable simulation. Therefore, a combined strategy utilizing both Veo and Genie could be beneficial for Google if they pursue development in this area.

Competitive Landscape

Google may face competition from companies like Microsoft, Scenario, Runway, Pika, and eventually, OpenAI with its Sora video-generating model.

Implications for the Industry

Considering Google’s planned advancements in world models and its history of leveraging substantial resources and distribution networks, competitors in this field should closely monitor Google’s progress.