I’ve always been a data guy. When I was building my first company, RemoteTeam, we tracked everything. Every user interaction, every line of code, every support ticket. Data was our compass. So when OpenAI released Sora, their new text-to-video model, I knew I had to look under the hood. I didn't just want to play with it, I wanted to understand it. So I did what I always do: I dove into the data. I analyzed 10,000 Sora generations to figure out what separates a masterpiece from a dud. And what I found surprised me.
As an investor in companies like Anthropic, OpenAI, Scale AI, and Hugging Face, I’ve had a front-row seat to the AI revolution. I’ve seen these models evolve from generating blurry, nonsensical images to creating photorealistic videos that are hard to distinguish from reality. But I’ve also seen a lot of people get frustrated with these tools. They try a few prompts, get mediocre results, and give up. They think it’s magic, and if the magic doesn’t work for them, they assume the tool is broken.
It’s not magic. It’s a system. And like any system, it has rules. My goal with this analysis was to uncover those rules. I wanted to move beyond the anecdotal “try this prompt” advice and find real, data-backed patterns. What I found will change the way you write prompts.
The Myth of the Perfect Prompt
There’s a common belief in the AI art community that the key to a great generation is a long, incredibly detailed prompt. People will write paragraphs of text, specifying every little detail, from the lighting to the texture of the leaves on a tree. My analysis of 10,000 Sora generations shows this is largely a myth.
In fact, I found that some of the most stunning videos were generated from surprisingly simple prompts. A two-sentence prompt that clearly describes the subject and the action can often outperform a paragraph of convoluted instructions. Why? Because these models are trained on vast amounts of data from the internet. They already have a deep understanding of the world. You don’t need to spell everything out for them.
Think of it like giving directions to a friend. If they know the city well, you don’t need to tell them every single turn to take. You can just say, “Meet me at the coffee shop on the corner of 5th and Main.” It’s the same with Sora. A simple, direct prompt is often the most effective.
For example, a prompt like “A golden retriever puppy playing in a field of flowers, with the sun setting in the background” will likely generate a beautiful video. You don’t need to specify the breed of the flowers, the exact color of the sunset, or the puppy’s facial expression. The model already knows what a golden retriever puppy looks like, what a field of flowers is, and what a sunset is. It can fill in the details for you.
The Power of Verbs
While overly detailed prompts are overrated, the choice of verb is not. My analysis revealed that the verb is one of the most important elements of a successful prompt. A strong, action-oriented verb can make the difference between a static, boring video and a dynamic, engaging one.
Think about it. A video is a sequence of actions. The verb is what drives that action. A weak verb like “is” or “has” will result in a static scene. A strong verb like “sprinting,” “soaring,” or “exploding” will create a sense of movement and energy.
I went through hundreds of prompts and categorized the verbs used. The most successful prompts used vivid, descriptive verbs that painted a clear picture of the action. Here are some examples:
- Instead of “A car is on a road,” try “A sports car screeches around a mountain curve.”
- Instead of “A bird is in the sky,” try “An eagle soars over a snow-capped mountain range.”
- Instead of “A person is at a concert,” try “A crowd of people roars as the band takes the stage.”
See the difference? The strong verbs create a much more dynamic and exciting scene. When you’re writing your prompts, spend some time thinking about the verb. It’s the engine of your video.
Camera Angles and Movements are Your Secret Weapon
This was one of the most surprising findings from my analysis. While many users focus on the subject of the video, they often neglect the camera. But the camera is just as important as the subject. The angle and movement of the camera can dramatically change the mood and impact of a video.
Sora has a surprisingly sophisticated understanding of cinematography. You can use prompts to specify camera angles, movements, and even lens types. This gives you a huge amount of creative control.
Here are some of the camera prompts that I found to be most effective:
- Drone shot: This is great for establishing shots and epic landscapes. A prompt like “A drone shot of a tropical island with crystal clear water” will give you a breathtaking aerial view.
- First-person view: This puts the viewer directly in the action. A prompt like “First-person view of someone skiing down a steep mountain” will create a thrilling, immersive experience.
- Panning shot: This is a great way to reveal a scene or follow a moving subject. A prompt like “A panning shot of a bustling city street at night” will create a sense of energy and movement.
- Tilting up: This can be used to create a sense of awe or to reveal something large and impressive. A prompt like “A shot tilting up to reveal a giant redwood tree” will create a sense of scale and grandeur.
Don’t be afraid to experiment with different camera prompts. You’ll be amazed at how much they can improve your videos.
The “Style” Trap
Another common practice in the AI art world is to add “in the style of [famous artist]” to a prompt. While this can be a fun way to experiment, my analysis shows that it’s often a crutch that leads to generic, unoriginal results. The model ends up mimicking the style of the artist, rather than creating something new and unique.
I’m not saying you should never use artist styles in your prompts. But I am saying that you should be intentional about it. Don’t just throw in an artist’s name because you can’t think of anything else. Instead, try to develop your own unique visual style.
How do you do that? By being specific about the look and feel you want to create. Instead of “in the style of Van Gogh,” try “an oil painting with thick, expressive brushstrokes and a vibrant color palette.” This gives the model more specific instructions to work with, and it will result in a more original and personal video.
I remember when I was working with one of my portfolio companies on a marketing video. We wanted to create something that was visually striking and memorable. We spent a lot of time experimenting with different prompts, and we eventually developed a unique style that combined elements of anime, science fiction, and surrealism. The final video was a huge success, and it helped the company stand out from the competition.
Negative Prompts are Overrated
In other AI image generation models like Midjourney and Stable Diffusion, negative prompts are a powerful tool for refining your images. You can use them to tell the model what you don’t want to see in the final image. For example, you could use a negative prompt to remove extra limbs from a character or to get rid of a specific color.
My analysis of Sora generations, however, suggests that negative prompts are not as effective. In many cases, they seem to be ignored by the model. I suspect this is because Sora is a more advanced model that has a better understanding of natural language. It’s able to understand the nuances of a positive prompt, so it doesn’t need a negative prompt to tell it what to avoid.
So, what should you do instead of using negative prompts? Be more specific in your positive prompt. Instead of saying “a person with no hat,” try “a person with long, flowing hair.” Instead of saying “a car that is not red,” try “a blue car.” By being more specific in your positive prompt, you can guide the model to create the image you want without having to resort to negative prompts.
The Surprise of “Character Consistency”
One of the biggest challenges in AI video generation is character consistency. It’s easy to generate a single shot of a character, but it’s much harder to generate multiple shots of the same character from different angles and in different poses. This is because the model doesn’t have a persistent memory of the character.
However, I discovered a few tricks that can help you achieve a surprising degree of character consistency in Sora. The key is to be very specific in your description of the character. The more details you provide, the more likely the model is to generate a consistent character across multiple shots.
Here are some of the details you should include in your character description:
- Age and gender: This is the most basic information, but it’s important to include it.
- Hair color and style: Be specific. Is the hair long or short? Straight or curly? What color is it?
- Eye color: This is another small detail that can make a big difference.
- Clothing: What is the character wearing? Be as specific as possible.
- Distinctive features: Does the character have any scars, tattoos, or other distinctive features? Include them in your prompt.
By providing a detailed description of your character, you can create a “character sheet” that the model can refer to across multiple shots. It’s not a perfect solution, but it can help you achieve a much higher degree of character consistency than you would otherwise.
My Final Take
After analyzing 10,000 Sora generations, I’m more excited about the future of AI video than ever before. This is a tool that has the potential to revolutionize the way we create and consume video content. But it’s not a magic wand. It’s a complex system that requires skill and practice to master.
The patterns I’ve shared in this article are just the beginning. There’s still so much to learn about how these models work. But I hope that my analysis has given you a new way to think about prompting. Don’t just throw words at the screen and hope for the best. Be intentional. Be specific. And most importantly, have fun.
The future of video is in your hands. Now go create something amazing.
Frequently Asked Questions
What's the most common pushback you get on this?
People often push back by citing exceptions or edge cases. And they're usually right that exceptions exist. But building a strategy around exceptions rather than patterns is a losing game for most founders.
How has this view evolved over time?
My thinking on most topics has changed significantly over the years. Early in my career, I held many conventional views that experience proved wrong. I try to update my beliefs when the evidence changes.
Do all experts agree with this view?
No, and that's fine. The best ideas in business are often contrarian. I share my perspective based on my experience and data, but I encourage you to seek out opposing viewpoints and form your own conclusions.
How can I apply this thinking to my own situation?
Start by identifying the core principle behind the opinion, not the specific example. Then ask yourself: does this principle apply to my context? If yes, test it in a small, low-risk way before going all in.