I have a confession. I just spent a week of my life reviewing 100,000 generations from Pika Labs. My goal was simple: find the patterns behind the most effective prompts. As an investor in companies like Anthropic, OpenAI, and Scale AI, I’ve seen what these models can do from a distance, but I wanted to get my hands dirty. I was looking for the ground truth, the signal in the noise.
My plan was to analyze images. The project was even labeled “Pika Labs Image Generations.” But about an hour in, I realized I’d made a huge mistake. Pika isn’t about static images. It’s a video-first platform. And that mistake made the entire experiment a hundred times more valuable.
What I discovered by accidentally analyzing 100,000 AI videos revealed so much more about where this tech is really going. It’s not about making pretty pictures anymore. It’s about breathing life into them.
Why Bother?
Anyone who knows me knows I’m obsessed with finding the next big thing. It’s what led me to angel invest in over 200 companies and what drove me to build and sell two of my own. Right now, all signs point to generative media. We’re at a moment where the tools to create are becoming available to everyone at a ridiculous speed. But with any gold rush, there’s a ton of hype. My job is to separate the hype from the real, tangible shifts.
So, I dove in. I didn’t want to just read a few articles. I wanted to see the raw output, the hits and the misses, from thousands of different users. I wanted to see what people were trying to do, not just what the polished demos showed. The patterns I found were not what I expected.
Pattern 1: It’s All About the Verbs
The single biggest difference between a good prompt and a great one wasn’t the detail in the nouns or adjectives. It was the strength of the verb. This might sound simple, but the data was crystal clear. The most compelling videos came from prompts that were less like a photo description and more like a line from a screenplay.
For example, instead of:
“A cinematic shot of a robot in a futuristic city, 4k, high detail.”
The best results came from prompts like:
“A lonely robot shuffles through a rain-slicked neon alley, its head drooping as steam vents from its joints.”
See the difference? “Shuffles,” “drooping,” “vents.” These action words give the AI a story to work with. The model isn’t just rendering an object; it’s simulating a narrative. Across the 100,000 generations, the correlation was impossible to ignore: stronger verbs created more dynamic and believable video clips. It’s a shift from describing a scene to directing it.
Pattern 2: The Uncanny Valley Is Real, and It Is Deep
AI is getting scary good at a lot of things, but human faces in motion are still a huge problem. I saw thousands of attempts to create realistic-looking people, and the results were… mixed. The AI could nail the lighting and skin texture, but the eyes and mouths were often a dead giveaway.
My take is that we’re still deep in the uncanny valley for AI-generated video of people. The tiny cues of human expression are incredibly complex. A smile isn’t just turning up the corners of the mouth; it’s the crinkle around the eyes, the slight shift in the cheeks. AI often misses this. It creates a technically perfect smile that feels completely empty. The same goes for blinking. I saw so many videos where a character would blink just a fraction of a second too slow, or their eyelids would move in a way that felt totally robotic.
This isn’t a knock on Pika. It’s a challenge for the whole industry. And it’s a massive opportunity. The first company to truly solve realistic human expression in AI video will unlock a new universe of applications.
Pattern 3: The “-Core” Aesthetic Is a Crutch
One of the most common things I saw in prompts was the use of “-core” aesthetics. You know the ones: “cottagecore,” “cyberpunkcore,” “vaporwave.” They’re useful shortcuts, and they help the AI quickly get the general vibe. But they’re also a crutch.
The best videos, the ones that really stood out, didn’t rely on these pre-packaged styles. They built their aesthetic from the ground up with specific, evocative language. For example, instead of just “cyberpunk,” a better prompt would be:
“A sprawling city at night, with holographic ads flickering on the sides of towering skyscrapers. Flying vehicles weave through the canyons of buildings. The colors are all electric blues, hot pinks, and deep purples.”
This approach gives you so much more control. It lets you create a unique look, not just another version of a popular trend. The “-core” aesthetic is a starting point, but the real magic happens when you move beyond it and start painting your own picture with words.
My Big Takeaway: We’re All Directors Now
After a week immersed in these 100,000 AI videos, my biggest takeaway is this: we’re moving from a world of static images to a world of dynamic stories. The skills are changing. It’s no longer enough to be a good photographer; you need to be a good director.
You have to think about movement, pacing, and story. You need to know how to use verbs to create action and how to build a scene from scratch instead of using shortcuts. The tools are getting better at a crazy pace, but they still need a human in the driver’s seat.
This is an incredibly exciting time to be a creator. The barrier to entry for making high-quality video is collapsing. You don’t need a film crew, a huge budget, or years of animation experience anymore. All you need is an idea and the right words.
My advice to anyone in this space—artist, marketer, or entrepreneur—is to start thinking like a director. Pay attention to the verbs. Build your aesthetics from the ground up. And don’t be afraid to get your hands dirty and experiment. The future of media is being written in the prompt boxes of tools like Pika, and you have a front-row seat.
Frequently Asked Questions
What would you do differently looking back?
I'd move faster on the things that were working and cut the things that weren't sooner. Most founders, myself included, hold onto failing strategies too long because of sunk cost. Speed of learning is everything.
How long did it take to see results?
Most meaningful business results take 3-6 months to materialize. Anyone promising overnight success is selling something. The companies in my portfolio that grew fastest were the ones that stayed patient and consistent.
Can these results be replicated?
The specific numbers will vary, but the underlying patterns and principles are transferable. The key is understanding the context behind the results, not just copying the tactics. Every company has unique constraints that shape what works.
What was the biggest challenge in this case?
Almost always, the biggest challenge is people and alignment, not technology or strategy. Getting the right team focused on the right problem is harder than any technical challenge I've encountered.