Google just launched Veo 3 this week, but it will be another week before I can get access to run my standard prompt tests on the new model. In the meantime, let's take a look back at Veo 2 and how it handled my benchmark prompts for video generation.

Veo 2 represented a significant step forward in AI video generation when it was released, and these results from my standard test prompts show both the capabilities and characteristics of Google's approach to video generation. As I wait for Veo 3 access, this serves as a good baseline to compare against what the newer model will deliver.

Here are the results using my standard prompt set that I use across different AI models for consistency in evaluation:

Castle Scene

The castle prompt typically tests how well a model handles architectural details and atmospheric elements in motion. Veo 2's approach to this classic fantasy scene shows the model's handling of complex environmental elements.

Family Portrait

Family scenes are crucial for testing how video models handle multiple human subjects and natural interactions. This result demonstrates Veo 2's approach to generating realistic human figures and group dynamics.

Mickey Character

The Mickey prompt is always interesting to see how different models handle copyrighted characters. Veo 2's interpretation provides insight into Google's approach to intellectual property concerns in their video generation models.

Messi Portrait

Celebrity prompts test how models handle recognizable public figures. The Messi prompt results show how Veo 2 navigates the challenge of generating content based on real people while managing potential legal and ethical considerations.

Robot Concept

Robot and mechanical subjects often showcase a model's ability to handle technical details and futuristic concepts. This result demonstrates Veo 2's capability with artificial and mechanical subject matter.

Looking Ahead to Veo 3

With Veo 3 just announced this week, it will be fascinating to see how Google has improved upon these foundational capabilities. Video generation models are advancing rapidly, and each iteration brings new capabilities in motion handling, temporal consistency, and overall video quality.

I'll be running this same prompt set on Veo 3 once I get access, which should provide a direct comparison to see how the technology has evolved. The improvements in video generation models have been dramatic over the past year, and Veo 3 promises to continue that trajectory.

These Veo 2 results serve as a good benchmark for where Google's video generation capabilities stood before their latest release. As we wait for broader access to Veo 3, these examples show the solid foundation that Google built with their previous generation model.