I almost gave up on AI image generation.
Seriously. I’d spent a small fortune on Midjourney and Stable Diffusion, burning through credits and getting… garbage. Funky hands, distorted faces, and images that looked nothing like what I had in my head. It felt like a scam. I’m a builder, an entrepreneur. I’ve had two successful exits, RemoteTeam and MovieLaLa, and I’ve invested in over 200 startups, including some of the biggest names in AI like Anthropic and OpenAI. I’m supposed to be good at this stuff. But I was failing. Miserably.
That frustration sent me down a rabbit hole. I decided I wasn’t just going to learn prompt engineering. I was going to master it. For the next six months, I treated it like a full-time job. I read every guide, watched every tutorial, and spent thousands of hours experimenting. I went from creating monstrosities to generating photorealistic images that have fooled professional photographers.
It was a grind. But you don’t have to go through that. I’m going to share the seven most important lessons I learned. These aren’t the generic tips you’ll find in-a-minute blog posts. These are the hard-won insights that will actually make a difference in your results.
1. Stop Being Polite. Be a Demanding Art Director.
My first mistake was treating the AI like a helpful intern. I’d write prompts like, “Could you please create a picture of a futuristic city?” The results were generic, boring, and looked like every other AI image on the internet.
The breakthrough came when I started acting like a demanding film director on a multi-million dollar set. I stopped asking. I started commanding.
Instead of “a futuristic city,” my prompts became brutally specific.
“An ultra-photorealistic, cinematic shot of a rain-slicked street in a cyberpunk city. Towering neon-lit skyscrapers with holographic ads. Flying vehicles whizzing by. Moody, atmospheric lighting, deep shadows, reflections on the wet pavement. Shot on a Sony A7S III with a 35mm f/1.4 lens. 8K resolution.”
See the difference? I’m not just describing a scene. I’m dictating the camera, the lens, the lighting, the mood, and the composition. I’m the boss. You have to be. These models have seen billions of images. You need to give them guardrails, or they’ll just give you the most average, statistically likely result.
2. The “Magic” is in the Modifiers
I used to focus all my energy on the subject. “A dragon.” “A knight.” “A castle.” But the real power isn’t in the noun. It’s in the adjectives, the adverbs, and the technical specifications. These are what I call “modifiers.”
Here’s a real example from one of my projects. I needed an image for a post about the future of work. My first prompt was simple: “A person working on a laptop in a futuristic office.” The result was a stock photo cliché.
Then I started layering modifiers.
- Style: “An impressionistic painting” vs. “A gritty, documentary-style photograph.”
- Artist: “In the style of Van Gogh” vs. “In the style of Annie Leibovitz.”
- Lighting: “Golden hour lighting” vs. “Harsh, fluorescent office lighting.”
- Composition: “Wide-angle shot” vs. “Extreme close-up.”
- Emotion: “A look of intense focus” vs. “A look of quiet desperation.”
My final prompt was something like: “A lonely entrepreneur, bathed in the glow of a laptop screen in a dark, minimalist apartment. A single cup of coffee on the desk. The city lights blur in the window behind him. A sense of quiet determination and isolation. Photorealistic, shallow depth of field.”
That prompt didn’t just create an image. It told a story. And it was all because of the modifiers.
3. Master the Art of the “Negative Prompt”
This one feels counterintuitive, but it’s a total game-changer. Most AI image tools have a feature called a “negative prompt.” It’s where you tell the AI what you don’t want to see. And it’s just as important as the main prompt.
AI models have weird biases. Ask for a picture of a doctor, and you’ll probably get a man. Ask for a beautiful landscape, and you might get ugly watermarks or text. The negative prompt is your weapon against this.
I always have a standard set of negative prompts I use for almost every image:
ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, extra limbs, disfigured, deformed, body out of frame, bad anatomy, watermark, signature, cut off, low contrast, underexposed, overexposed, bad art, beginner, amateur, distorted face
This cleans up so many of the common AI issues. Then, I add specific negative prompts based on the image. If I’m getting too many people in a shot, I’ll add crowd to the negative prompt. If the colors are too garish, I’ll add saturated. It’s like sculpting. You’re not just adding clay; you’re also carving it away.
4. Numbers Are Your Friend
This is a lesson I learned from my investing career. Vague statements are useless. Specific numbers get results. It’s the same with prompting.
Don’t say “a few people.” Say “three people.”
Don’t say “tall buildings.” Say “100-story skyscrapers.”
I was trying to create an image of a complex circuit board once. My prompts were getting me messy, chaotic results. Then I tried specifying the number of components. “A circuit board with 12 capacitors, 8 resistors, and 1 large central processing unit.” The difference was night and day. The AI suddenly understood the structure I wanted.
This works for everything. The number of trees in a forest. The number of ships in a fleet. The number of wrinkles on a face. Be precise. The AI will reward you for it.
5. It’s an Iterative Conversation, Not a Vending Machine
I think the biggest mental shift for me was realizing that prompt engineering isn’t a one-shot process. You don’t just put in a prompt and get a perfect image. It’s a back-and-forth conversation.
You write a prompt. You get a result. You analyze what worked and what didn’t. You refine the prompt. You try again. And again. And again.
My workflow for a single image can involve 20 or 30 different prompt variations. I’ll start with a simple idea and then build on it. I’ll try different lighting, different camera angles, different styles. I’ll use features like “vary region” to fix specific parts of the image that I don’t like.
This is where most people give up. They try a couple of times, don’t get what they want, and blame the tool. That’s like a painter blaming the brush. You have to be willing to put in the work. The AI is a partner in the creative process, not a magic lamp.
6. Steal Like an Artist (But Be Smart About It)
I’m not telling you to plagiarize. But I am telling you to learn from the best. There are communities online where people share their prompts. Websites like Lexica and OpenArt are treasure troves of inspiration.
When I see an image I love, I don’t just admire it. I dissect it. I break down the prompt. What modifiers did they use? What artists did they reference? What technical specifications did they include?
I’ll take that prompt and then make it my own. I’ll swap out the subject. I’ll change the color palette. I’ll combine it with ideas from other prompts. This is how you develop your own unique style. You’re not copying; you’re learning the language. You’re building your vocabulary.
I’ve learned more from reverse-engineering other people’s prompts than from any tutorial. It’s like looking at the source code of a great piece of software. The secrets are all there, hiding in plain sight.
7. The Future is Multimodal
The world of AI is moving incredibly fast. Just a year ago, we were struggling to make hands that didn’t look like spaghetti. Now, with models like Sora, we’re generating entire video scenes from a single text prompt.
This is the final, and most important, lesson. Don’t just think in terms of text-to-image. The future is multimodal. It’s text-to-video, image-to-image, text-to-3D-model. The skills you’re learning in prompt engineering for still images are the foundation for all of it.
The ability to translate a creative vision into a precise, detailed, and effective prompt is going to be one of the most valuable skills of the next decade. It’s the new literacy.
I’ve used these techniques to create everything from marketing assets for my portfolio companies to concept art for a sci-fi project I’m working on. It has saved me tens of thousands of dollars in freelance fees and countless hours of frustration.
So, don’t give up. Don’t be intimidated. Start with these seven lessons. Be demanding. Be specific. Be iterative. And start building the future. You’ll be amazed at what you can create.
Frequently Asked Questions
Which item on this list has the highest impact?
It depends on your stage and context, but in my experience, the items near the top of the list tend to have the broadest applicability. That said, sometimes the less obvious items create the biggest breakthroughs for specific situations.
Can I implement all of these at once?
I'd strongly recommend against it. Pick the 2-3 items that resonate most with your current situation and focus there. Trying to do everything simultaneously is a recipe for doing nothing well.
How do I know which items apply to my situation?
Start by honestly assessing where your biggest bottleneck is right now. The items that address that specific constraint will give you the highest return on your time and energy.