There’s a new king of AI video, but it comes with a huge catch.
I was scrolling through my feed when I found this massive deep dive into the latest AI video models, and I had to share it. This is one of the most thorough comparisons I’ve ever seen! The creator put three of the newest, most powerful models, Veo 3.1, Sora 2 Pro, and Wan 2.5, through an insane gauntlet of tests. The results are seriously surprising.
The expert’s goal was to figure out which model is best for which specific task. He tested everything from complex physics and emotional dialogue to full-on musical performances and unique art styles. While one model did come out on top overall, the real takeaway is that each one has a distinct superpower you need to know about.
Here’s the breakdown of what the mind behind it discovered:
📌 Physics & Action Scenes
Sora 2 Pro achieved something incredible. It was the only model that could successfully generate a video of dominoes falling correctly, a challenge that has stumped AI for ages. The author also found it was generally the strongest for complex movements like running and fight scenes, even if it had some minor warping issues.
💡 Music & Audio Generation
When it came to audio, Veo 3.1 absolutely stole the show. This industry pro showed how it could generate a genuinely catchy song from a prompt about a singing frog, and even created a full pop-punk band performance where the headbanging synced to the music. For anything involving music, Veo is the undisputed champ.
✅ Dialogue & Unique Features
Dialogue was a much closer race. The most interesting finding from the person who shared it was how the unique features made a difference. Veo 3.1’s new “start-and-end-frame” function is a filmmaking beast, allowing for shots that were previously impossible. Meanwhile, Wan 2.5 was the winner for maintaining complex art styles from images. Sora was great at authentic, emotional dialogue, but its inability to use realistic faces in image-to-video prompts is a major drawback.
This is just scratching the surface. The original post is packed with dozens of side-by-side video examples that you have to see for yourself. Check out the full analysis to see all the clips and decide which tool is right for your next project.