I’ve covered a wide range of artificial intelligence tools over the years including video, music, code and image generators. One of the most fun, and most frustrating is AI video. It is changing and improving so fast it is difficult to track the best approach to creating content.

I’ve learned that sometimes the simplest tests reveal the most telling results. That’s why I decided to put the latest crop of AI video generators through their paces with some straightforward scenarios.

The current landscape is dominated by some fascinating tools — Runway Gen 3, Kling 1.6, and Pika Labs new 2.1 model are among the best. But they’re not close to the full picture as we’ve also got models from Haiper, Luma Lab’s Ray2 and Hailuo’s MiniMax. That doesn’t cover the open source options.

Rather than throwing complex scenarios at them, I opted for something more fundamental. Why? Because in my experience, even the most advanced AI can stumble over basic tasks when multiple elements compete for attention.

My Test Scenarios

I crafted two simple scenes that any competent video tool should handle covering ordinary people doing very ordinary things. I would have done more but Kling generations taken too long.

A coffee artisan at work — specifically focusing on a barista creating latte art. It’s a single action that requires fluid motion and attention to detail.

A moment of quiet contemplation — capturing someone absorbed in a book beside a rain-streaked window. Simple in concept, but requires subtle atmospheric elements.

The Contenders

I ran these prompts through five leading platforms: Runway Gen 3-Alpha, Kling 1.6, Pika Labs 2.1, Luma Labs Ray2, and Hailuo MiniMax. Each brought something unique to the table.

The Coffee Test

When I prompted “A barista carefully pouring steamed milk into a ceramic cup, forming a delicate heart-shaped latte art. The warm light of the café softly illuminates the counter,” the results were revealing.

Hailuo MiniMax

Kling 1.6

Pika 2.1

Runway Gen-3

Luma Labs Ray2

The Reading Scene

For the second test, I used: “A woman sitting by a large rain-streaked window, holding an open book in her lap. She occasionally glances outside at the overcast sky, lost in thought.“

Hailuo MiniMax

Kling 1.6

Pika 2.1

Runway Gen-3

Luma Labs Ray2

The Verdict

Having worked with various AI tools over the years, what strikes me most is how all five models did much better with the second, woman in the rain prompt that then pouring coffee. I was also impressed with how different each model performed with unique takes on the same prompt.

These results underscore a crucial point for creators: sometimes less is more. While these tools are advancing rapidly, they’re still at their best when handling focused, single-action scenes.

The future of AI video generation looks incredibly promising, but for now, I’d recommend keeping things simple and playing to each tool’s strengths. As someone who’s watched this space evolve, I’m fascinated to see where we’ll be in another six months.