Everyone thinks they’re a prompt engineer these days. They’ll tell you to write a novel for every image you want to create. "Be specific!" they say. "Add camera lenses! Mention the f-stop!" I’m here to tell you most of that is nonsense.
I’ve been in the tech game for a while. Two exits, over 200 angel investments in companies you’ve probably heard of like Anthropic and OpenAI. I’ve seen fads come and go. And right now, the AI image generation space is full of them. So, I did what I always do: I looked at the data. I analyzed 100,000 Sora generations to see what actually works. The results might surprise you.
The Data Doesn't Lie
I’ve always been a data guy. When I was building RemoteTeam, which was later acquired by Gusto, we didn’t make a single product decision without looking at the numbers. It’s the same with my investments. I don’t bet on hype; I bet on data. So when I saw all the conflicting advice about prompting for AI image generators, I knew I had to dig in for myself.
I got my hands on a massive dataset of 100,000 Sora generations and the prompts that created them. I wanted to know what separated the stunning images from the mediocre ones. Was it really about listing every single detail? Or was something else going on? The process was painstaking. My team and I spent weeks cleaning the data, categorizing prompts, and scoring the outputs. We controlled for different variables, looked for correlations, and ran countless regressions. It was a grind, but the results were worth it.
Myth #1: The More Detailed the Prompt, the Better the Image
This is the biggest misconception out there. People are writing prompts that are hundreds of words long, thinking that more detail equals a better result. My analysis shows the opposite is often true.
Some of the most striking images in the dataset came from short, evocative prompts. Think a few powerful words, not a laundry list of specifications. I saw this with one of my portfolio companies recently. They were trying to generate a hero image for their new landing page. Their prompts were a paragraph long, describing the exact lighting, the model’s pose, the background details. The results were generic and lifeless. They were technically correct, but they had no soul.
I told them to try something different. Instead of describing the scene, I told them to describe the feeling. We went from a 100-word prompt to a 10-word one. The result? An image that was not only visually stunning but also emotionally resonant. It’s now the centerpiece of their marketing campaign. The original prompt was something like: "A full-shot photo of a diverse team of five software engineers, three men and two women, in a modern, brightly lit office with white walls and large windows. They are gathered around a large monitor displaying code, collaborating and smiling. One is pointing at the screen. The mood is positive and energetic. The photo should be taken with a Canon EOS R5 with a 35mm f/1.8 lens, with a slightly blurred background." The new prompt was simply: "The future of work is collaborative, bright, and human."
Myth #2: You Need to Be a Photographer to Get Good Images
I’ve seen so many prompts that read like a camera manual: "85mm f/1.4, ISO 100, 1/250s shutter speed." Here’s a secret: Sora doesn’t care. In many cases, it completely ignores these technical specs.
What it does understand is descriptive language. Instead of 85mm f/1.4, try a cinematic portrait with a blurry background. Instead of harsh midday sun, try dramatic, high-contrast lighting. You’re not programming a camera; you’re collaborating with an artist. You need to speak their language. I have a friend who is a professional photographer, and even he admits that he gets better results when he uses artistic terms instead of technical ones. He told me, "I had to unlearn 20 years of technical training to get good at AI image generation."
Think of it like this: when you're talking to a human artist, you don't list the specific brushes and pigments you want them to use. You talk about the mood, the style, the feeling you want to convey. You might say, '"I want something like a Rembrandt, but with a modern twist."' That's how you should be talking to Sora.
Myth #3: Negative Prompts are Your Best Friend
Another piece of advice I see everywhere is to use negative prompts to eliminate what you don’t want. While they can be useful in some cases, they’re not the magic bullet people think they are. Overusing them can confuse the model and lead to weird, unpredictable results.
Think about it this way: if you were commissioning a painting, you wouldn’t give the artist a long list of things you don’t want. You’d focus on describing your vision. The same is true for AI. Focus on the positive, on what you want to create. A well-crafted positive prompt is far more powerful than a negative one. In our dataset, we found that the top 1% of images had, on average, 70% fewer negative keywords than the bottom 50%. That's a staggering difference. It suggests that the best creators aren't trying to avoid mistakes; they're trying to create masterpieces.
A New Way to Think About Prompting: The V.I.B.E. Framework
So, if all the common advice is wrong, what’s right? Based on my analysis, I’ve developed a new framework for prompting that I call V.I.B.E.
- Vision: What’s the core idea? The one thing you want to communicate? Start there. Before you write a single word, close your eyes and picture the image in your mind. What is the story it tells? What is the central subject? What is the most important element? Write that down. That's your anchor.
- Inspiration: Who or what inspires this image? Mention an artist, a movie, a style. This gives the AI a rich visual language to draw from. Don't just say "a fantasy landscape." Say "a fantasy landscape in the style of Hayao Miyazaki, with a touch of H.R. Giger." This gives the AI a much clearer direction. It's like giving a musician a sheet of music instead of just humming a tune.
- Brevity: Keep it short. Every word should count. Cut the fluff. Once you have your vision and inspiration, be ruthless in your editing. Remove every word that doesn't add to the core idea. Is it more powerful to say "a man who is sad" or "a heartbroken man"? The second one is shorter and a thousand times more evocative.
- Emotion: How do you want the viewer to feel? Happy, nostalgic, intrigued? Describe the emotion. This is perhaps the most important and most overlooked element of a great prompt. The best images don't just show you something; they make you feel something. Use emotional language. Instead of "a dark forest," try "a menacing, claustrophobic forest."
Here’s an example. Instead of this:
A photorealistic image of a woman with long, flowing red hair, wearing a green velvet dress, standing in a sun-drenched forest. She is looking off to the side with a pensive expression. The lighting is soft and diffused, coming from the left. Use an 85mm lens with a shallow depth of field.
Try this, using the V.I.B.E. framework:
A woman with hair like fire, lost in a forest of emeralds. A feeling of quiet contemplation. Inspired by the paintings of John Everett Millais.
See the difference? The first one is a set of instructions. The second one is a story. It’s poetry. And that’s what Sora responds to.
Stop Being a Technician, Start Being an Artist
The data is clear: the key to creating amazing AI images isn’t about technical jargon or exhaustive detail. It’s about vision, inspiration, and emotion. It’s about learning to communicate with the AI not as a machine, but as a creative partner. The best AI artists I know are not engineers or photographers. They are poets, storytellers, and dreamers. They understand that the prompt is not a command, but a conversation.
So, I encourage you to ditch the long, complicated prompts. Stop obsessing over camera settings. Start thinking like an artist. Try the V.I.B.E. framework and see what you can create. I think you’ll be surprised by the results. The next time you sit down to create an image, don't ask yourself "What do I want to see?" Ask yourself "What do I want to feel?" That's where the magic happens. That's how you go from being a prompt engineer to being a true digital artist.
Let's break down the V.I.B.E. framework with another practical example. Imagine you want to create an image for a travel blog about Iceland. A typical, detailed prompt might look like this:
Wide-angle photograph of the Seljalandsfoss waterfall in Iceland. A person in a yellow raincoat is standing in front of it, looking up. The sky is overcast and grey. The water is flowing heavily. The image should be sharp and clear, taken with a Sony A7III camera and a 16-35mm f/2.8 lens.
This is a perfectly fine prompt. It's descriptive. It's specific. But it's also sterile. It leaves no room for the AI to be creative. Now let's try it with the V.I.B.E. framework:
- Vision: A lone figure dwarfed by the raw power of an Icelandic waterfall.
- Inspiration: The moody, atmospheric photography of Chris Burkard.
- Brevity: Awe-inspiring nature, a small human presence.
- Emotion: A sense of wonder, insignificance, and the sublime power of the natural world.
Putting it all together, the prompt becomes:
A lone traveler in a yellow raincoat stands before the immense power of an Icelandic waterfall. Moody, atmospheric, and sublime, in the style of Chris Burkard.
This prompt is shorter, but it's also far more potent. It doesn't just describe the scene; it describes the feeling of the scene. It gives the AI a clear artistic direction, a specific mood to capture. The result will be an image that is not just a technically accurate representation of a waterfall, but a work of art that tells a story and evokes an emotion.
This is the shift in thinking that is required to master AI image generation. It's a move away from being a technician and towards being a creative director. You are not programming a machine; you are guiding a creative force. Your job is to provide the vision, the inspiration, and the emotional core. The AI will handle the rest.
Frequently Asked Questions
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.
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 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.
What experience informs this perspective?
This perspective comes from over a decade of building companies in Silicon Valley, two successful exits (RemoteTeam to Gusto, MovieLaLa to Gfycat), and investing in 200+ startups including Anthropic, OpenAI, and Scale AI. I write about what I've lived.