I’m about to say something that might get me kicked out of the prompt engineering club: most of what you’ve been told about writing good Midjourney prompts is probably wrong. Dead wrong.
For months, I’ve been seeing the same advice everywhere. "Be more descriptive." "Add more details." "Use complex sentences." I tried it all. My prompts became miniature novels. And the results? Sometimes they were great, but often they were a mess of over-stylized, generic-looking images that looked like everything else on the platform. It felt like I was just throwing spaghetti at the wall and hoping something would stick. It reminded me of the early days of my first startup, RemoteTeam. We were trying to build a product for a market that didn't exist yet. We were just guessing. And we were failing. A lot. That experience taught me a valuable lesson: when you're lost, you need a map. And the best map is data.
As an engineer and investor, I have a simple rule: when in doubt, follow the data. I’ve been fortunate enough to invest in some of the most exciting AI companies out there, like Anthropic, OpenAI, and Scale AI. I’ve seen firsthand how powerful data can be in cutting through the noise and revealing what actually works. So, I decided to apply the same thinking to Midjourney.
I got my hands on a dataset of 10,000 Midjourney generations, complete with the prompts that created them and user ratings for each image. I spent a week crunching the numbers, running statistical analyses, and looking for patterns. What I found was, to put it mildly, surprising. The data told a story that was completely different from the popular narrative. It turns out that the secret to creating amazing images with Midjourney isn't about writing longer prompts or adding more details. It's about being smarter with your words.
The Great Detail Myth
One of the first things I noticed was that there was no correlation between the length of a prompt and the quality of the generated image. In fact, some of the highest-rated images in the dataset were generated from incredibly simple prompts. We're talking 5-10 words.
For example, the prompt "a lone astronaut on a red desert" produced a stunningly evocative image that was rated in the top 1% of the entire dataset. Meanwhile, a much longer prompt like "a highly detailed, photorealistic image of an astronaut in a futuristic spacesuit, standing on the desolate, red-sanded surface of Mars, with a backdrop of distant, alien mountains and a sky filled with two moons, rendered in a cinematic style with dramatic lighting" produced a decent, but far more generic, image. Why? Because the first prompt gives Midjourney a clear, powerful concept to work with. The second prompt bombards it with so many details that it doesn't know what to focus on. It's like trying to paint a masterpiece with a thousand different colors at once. You just end up with a muddy mess.
My analysis showed that prompts with 10-15 words had the highest average rating. After that, the ratings started to decline. It seems that once you give Midjourney too much information, it starts to get confused and averages things out, leading to less impactful images. The key is to give it just enough information to set the scene and then let it do its thing. Think of it like this: you're the director, not the micromanaging producer. Your job is to provide a clear vision, not to dictate every single detail.
The Power of Weird Words
Another fascinating pattern I discovered was the power of unusual or unexpected words. I ran a frequency analysis on the prompts and found that a small number of uncommon words had a disproportionately high correlation with top-rated images. Words like "iridescent," "bioluminescent," "chromatic," and "opalescent" were pure gold.
Think about it. These words don't just describe a color; they describe a quality of light and texture that is inherently interesting and visually complex. When you use a word like "iridescent," you're not just telling Midjourney to make something shiny; you're giving it a whole concept to play with. The results are often surprising and beautiful in a way that a simple color description could never be. It's the difference between saying "a blue car" and "a car the color of a peacock's feather."
I also found that certain artistic and technical terms, even if they're not commonly used, can have a huge impact. Words like "isometric," "voxel," and "chiaroscuro" were consistently associated with high-quality images. This suggests that Midjourney has a deep understanding of artistic principles and that you can use this to your advantage. Don't be afraid to raid art history books and technical manuals for inspiration. You might find that the perfect word for your next masterpiece is hiding in plain sight.
Camera and Lighting: The Unsung Heroes
If there's one thing you should be adding to your prompts, it's not more adjectives; it's camera and lighting directions. My analysis showed that prompts that specified a camera angle (e.g., "low-angle shot," "drone shot," "macro shot") and a lighting style (e.g., "Rembrandt lighting," "dappled light," "backlit") had a significantly higher chance of producing a top-rated image.
This makes perfect sense. Camera and lighting are the two most important tools a photographer has for creating a mood and telling a story. The same is true for AI image generation. By specifying the camera and lighting, you're not just describing a scene; you're directing it. You're telling Midjourney how you want the viewer to feel. A low-angle shot can make a subject feel powerful and imposing. A high-angle shot can make them feel small and vulnerable. Rembrandt lighting can create a sense of drama and intimacy. Dappled light can evoke a feeling of peace and tranquility.
Here's a simple example. The prompt "a portrait of a woman" is a lottery. You could get anything. But the prompt "a portrait of a woman, Rembrandt lighting, low-angle shot" is a recipe for a dramatic and powerful image. You're guiding the AI to create something with a specific emotional tone. You're no longer just a writer of prompts; you're a virtual photographer.
Case Study: From Boring to Brilliant
Let's put this all together with a real-world example. I started with a generic prompt that I saw someone use on a Discord server: "a fantasy castle in a forest." The results were... fine. They were castles. They were in forests. But they were also incredibly boring. They looked like something you'd see on the cover of a bad fantasy novel from the 1980s.
So, I decided to apply my "Concept, not Clutter" framework. I started with a new core subject: "a forgotten temple." Then, I added a conceptual keyword: "bioluminescent." Finally, I set the scene with camera and lighting: "drone shot, misty morning light."
The new prompt was: "a forgotten temple, bioluminescent, drone shot, misty morning light."
The difference was night and day. The new image was breathtaking. It was a stunningly original vision of a lost world, filled with mystery and wonder. The bioluminescent moss and fungi cast an ethereal glow on the ancient stones. The drone shot gave it a sense of scale and grandeur. The misty morning light added a touch of atmosphere and mood. It was a world I wanted to explore. And it was all because I stopped trying to describe everything and started focusing on a single, powerful concept.
A New Framework for Prompting
Based on my analysis, I've developed a new framework for writing Midjourney prompts that I call "Concept, not Clutter." The idea is simple: instead of cluttering your prompts with unnecessary details, focus on a single, powerful concept.
Here's how it works:
- Start with a core subject. This should be simple and to the point. "A lone astronaut," "a futuristic city," "a mythical creature."
- Add a conceptual keyword. This is where you use one of those "weird words" I talked about earlier. "An iridescent jellyfish," "a bioluminescent forest," "a chromatic aberration."
- Set the scene with camera and lighting. This is where you direct the shot. "Low-angle shot, Rembrandt lighting," "drone shot, golden hour," "macro shot, backlit."
That's it. Three simple steps. No long sentences. No laundry list of adjectives. Just a clear, concise concept that gives Midjourney a strong creative direction.
I've been using this framework for a few weeks now, and the results have been incredible. My images are more consistent, more creative, and more aligned with my vision. I'm spending less time writing prompts and more time exploring new ideas. It's a liberating feeling. It's like I've finally cracked the code.
I know this goes against a lot of the conventional wisdom out there. But the data doesn't lie. If you want to take your Midjourney creations to the next level, I urge you to give this a try. Forget everything you've been told about prompt engineering and start thinking like a concept artist. You might be surprised by what you create. The best part? You'll be creating more original work, not just another pretty picture that looks like everything else. You'll be creating your own worlds. And that, my friends, is where the real magic happens.
Frequently Asked Questions
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.
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.
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.