99 Lessons I Learned About AI-driven Design After Spending the last 2 years

I went deep on AI-driven Design, investing the last 2 years to master it. These are the 99 most critical, non-obvious lessons that will accelerate your learning curve and save you from costly mistakes.

I still remember the first time I tried using Midjourney. It was two years ago, and the hype around AI image generation was just starting to explode. I’d seen all the incredible images on Twitter and thought, "This is it. This is the future of design." I jumped in, typed a few prompts, and got… garbage. Absolute, unusable garbage.

It was a humbling experience. Here I was, a guy who’d built and sold two tech companies and invested in over 200 startups, including some of the biggest names in AI like Anthropic and OpenAI, and I couldn’t get a decent image out of a machine. It felt like trying to communicate with an alien who spoke a language I didn’t understand. I was frustrated, but I was also intrigued. I knew there was something powerful here, and I was determined to figure it out.

So, I went deep. I spent the last two years completely immersing myself in the world of AI-driven design. I’ve probably spent more than 10,000 hours experimenting with Midjourney, DALL-E, and a dozen other tools. I’ve generated hundreds of thousands of images. I’ve talked to the founders of the biggest AI companies. I’ve made a ton of mistakes, and I’ve learned a ton of lessons. And now, I want to share them with you. These are the 99 lessons that will save you from the frustration I went through and help you master AI-driven design.

Mindset & First Steps

Embrace the chaos. AI is not a precise tool. It's a creative partner. You have to be willing to let go of some control and see where it takes you.
Your first 100 images will suck. Mine did. Everyone's do. Don't get discouraged. It's a learning process.
Don't blame the tool. If you're not getting what you want, it's probably your prompt. Or your expectations.
Start with one tool and master it. I see too many people jumping between Midjourney, DALL-E 3, and Stable Diffusion. Pick one. Learn its nuances. For me, that was Midjourney.
Read the manual. Seriously. The Midjourney documentation is actually good. You’ll learn about parameters and commands that will save you hours.
Join the Discord community. It's the best way to see what other people are creating and how they're doing it.
Stop thinking in keywords. Think in sentences. Describe the image you want as if you were talking to a human artist.
Use negative prompts from day one. Telling the AI what you don't want is just as important as telling it what you do want.
Don't get obsessed with photorealism. It's impressive, but it's not the only thing AI can do. Explore different styles.
It's not about 'art', it's about communication. You're using images to tell a story, sell a product, or explain an idea. The image itself is just the medium.

Prompting: The Core Skill

Be specific. Insanely specific. Don't just say "a car." Say "a vintage 1967 Ford Mustang, candy apple red, parked on a rain-slicked street in Tokyo at night, with neon signs reflecting on the hood."
Use adjectives. Lots of them. Words like "cinematic," "dramatic," "ethereal," "gritty," and "vibrant" can completely change the mood of an image.
Reference artists and styles. Want a comic book look? Add "in the style of Jack Kirby." Want something painterly? Try "in the style of Rembrandt."
Don't forget the camera. I often include camera specs in my prompts. "Shot on a Canon 5D Mark IV with a 50mm f/1.8 lens" gives you a specific depth of field and look.
Chain your thoughts. Start with a simple prompt, see what you get, and then add to it. It's an iterative process.
The --chaos parameter is your friend. It introduces randomness and can lead to happy accidents. I use it all the time.
Use --stylize to control the 'artistic' level. A lower value is more literal, a higher value is more creative. Find the sweet spot for your needs.
Learn to use image prompts. Sometimes the best way to get what you want is to show the AI an image and have it riff on that.
Multi-prompting is a power move. You can combine concepts like "a cat::2 and a dog::1" to control the weight of each element.
Keep a prompt journal. I have a huge text file with all my best prompts. It's my secret weapon.

Workflow & Tools

Don't generate one image. Generate hundreds. The cost of generation is close to zero. The cost of a bad idea is high. Explore the solution space.
Use a dedicated folder for your generations. It gets messy, fast. I have a folder for each project.
Upscale everything. The default resolutions are often too small for professional use. I use Topaz Gigapixel AI to upscale my best images.
Photoshop is still your best friend. AI is a starting point, not the finish line. I take almost every image into Photoshop for cleanup, color correction, and compositing.
Learn the basics of photo editing. You don't need to be a Photoshop guru, but you need to know how to use layers, masks, and adjustment layers.
The clone stamp is your secret weapon for fixing AI weirdness. Weird hands? Extra fingers? The clone stamp can fix it in seconds.
Don't be afraid to composite multiple images. I often take the head from one generation, the body from another, and the background from a third.
Use a consistent seed for variations. If you find a style you like, use the same seed number to generate variations with different subjects.
The --tile parameter is amazing for creating seamless patterns. I've used this for website backgrounds, textiles, and more.
Don't sleep on the "Vary (Region)" feature. It's a game-changer for fixing specific parts of an image without re-rolling the whole thing.
Create a style library. I have a set of prompts that define my personal "brand" style. I can apply it to any new image.
Use ChatGPT to generate prompts. Sometimes I'll describe an image to ChatGPT and ask it to write a detailed Midjourney prompt for me. It's surprisingly good.
Krea.ai is a fantastic tool for upscaling and enhancing images. It has some unique features that other tools don't.
Magnific.ai is another incredible upscaler. It can add incredible detail to your images.
Don't ignore the built-in upscalers. Midjourney's own upscalers are getting better and better.
Your workflow should be: Idea -> Prompt -> Generate -> Select -> Upscale -> Edit -> Final. Don't skip steps.
Batch your work. I'll spend a few hours just generating images, then a few hours editing. It's more efficient.
Take breaks. Staring at AI images for hours can make you go a little crazy. Step away from the screen.
Don't fall for every new tool that comes out. There's a new AI tool every week. Stick with the ones that work for you.
The best tool is the one you know how to use. A master in Midjourney will get better results than a novice in a dozen different tools.

Advanced Techniques

Character consistency is the holy grail. It's still hard, but using the --cref parameter with a reference image is the best way to do it.
Style references (--sref) are more powerful than you think. You can use them to create a consistent look and feel across a whole series of images.
Use weird weights. Don't just stick to whole numbers. Try cat::1.2 dog::-0.8.
Permutation prompts are a power user's dream. You can generate dozens of variations with a single prompt, like {red, green, blue} car.
The --repeat parameter is great for exploring a prompt. I'll often run a prompt 10 or 20 times to see the full range of possibilities.
Remix mode is your creative co-pilot. It lets you change the prompt on a variation, which is perfect for refining an image.
Don't be afraid to use low --stylize values. Sometimes you want the AI to be more literal, especially if you're using a detailed prompt.
Use the /describe command to learn from other images. See an image you like? Use /describe to get four possible prompts that could have created it.
Aspect ratios matter. A 16:9 image will have a different composition than a 1:1 image. Think about your final output.
The --weird parameter is... weird. But it can lead to some truly unique and unexpected results. Use with caution.
Think about lighting. I often include lighting cues in my prompts, like "dramatic backlighting" or "soft morning light."
And color. A limited color palette can be very effective. Try "monochromatic blue" or "duotone red and black."
Composition is still king. Use the rule of thirds, leading lines, and other classic composition techniques in your prompts.
Negative space is your friend. Don't try to fill every inch of the frame.
Texture is everything. Add words like "rough," "smooth," "metallic," or "furry" to your prompts.
Don't just describe the subject, describe the emotion. "A sad robot" is a better prompt than "a robot."
Use analogies. "A city that looks like a circuit board" is a great way to get a unique image.
Combine unexpected elements. "A jellyfish made of glass" or "a treehouse on the moon."
The more you know about art history, the better your prompts will be. Knowing the difference between Baroque and Rococo will give you a huge advantage.
Break the rules. Once you know the rules, you can start to break them. That's where the real magic happens.

AI Video: The Next Frontier

AI video is where image generation was two years ago. It's clunky, weird, and full of artifacts. But it's getting better, fast.
Start with RunwayML. It's the most mature text-to-video platform right now.
Keep your prompts short and simple. Video models are not as sophisticated as image models yet.
Think in shots, not scenes. You're not going to generate a full movie. You're going to generate a 4-second clip.
Use image-to-video. The best results I've gotten have been from animating a still image I generated in Midjourney.
Embrace the surreal. AI video is great at creating dreamlike, abstract visuals. Don't fight it.
Pika Labs is another one to watch. They're doing some really interesting things with video generation.
Don't expect character consistency. It's even harder in video than it is in images.
Use it for B-roll and abstract backgrounds. That's the most practical use case right now.
The pace of improvement is staggering. What's impossible today will be easy in six months. Start learning now.

The Business of AI Design

This is a new profession. I call it "AI-driven Design." It's a hybrid of art direction, copywriting, and technical skill.
You can make a lot of money with this skill. I know people who are charging thousands of dollars for a single image.
Build a portfolio. No one will hire you if they can't see your work. Create a website or a Behance profile and show off your best stuff.
Find a niche. Don't try to be everything to everyone. Are you the go-to person for fantasy book covers? Sci-fi concept art? Architectural visualization?
Your value is not in pushing the button. It's in your taste, your ideas, and your ability to translate a client's vision into a compelling image.
Don't sell yourself short. Just because it only took you 10 minutes to generate an image doesn't mean it's not valuable. You're being paid for your years of experience and your thousands of hours of practice.
Educate your clients. Most people still don't understand what AI can and can't do. It's your job to explain it to them.
Over-deliver. Give your clients more than they asked for. Give them options. They'll love you for it.
The market is getting crowded. The only way to stand out is to be better than everyone else. That means putting in the work.
This is not a replacement for human designers. It's a tool. A very powerful tool, but still a tool. The best results will always come from a human and an AI working together.

Future & Predictions

Real-time generation is coming. You'll be able to see the image change as you type the prompt.
3D generation is the next big thing. We're already seeing the early stages of it with tools like Luma AI.
AI will be integrated into every creative tool. Photoshop, Figma, and even Google Slides will have powerful AI features built-in.
The prompt will become a new kind of code. We'll have prompt engineers, prompt optimizers, and prompt marketplaces.
Personalized models will be the norm. You'll be able to train an AI on your own style and have it generate images that look like they were made by you.
The uncanny valley will disappear. AI-generated humans will be indistinguishable from real ones.
We'll see the first AI-directed movie in the next 5 years. It might not be good, but it will happen.
The legal and ethical issues are going to get a lot more complicated. We need to have a serious conversation about copyright, ownership, and the impact on artists.
This technology will create more jobs than it destroys. But they will be different kinds of jobs. You need to adapt.
The pace of innovation will not slow down. If you think things are crazy now, just wait.

Final Thoughts

Stay curious. The moment you think you've mastered this, something new will come along and change everything.
Be patient. This is a marathon, not a sprint.
Don't be afraid to be weird. The most interesting results come from the most unexpected prompts.
Share your work. The community is a huge part of what makes this so exciting. Share your successes and your failures.
Teach others. The best way to learn is to teach. Help someone who is just starting out.
Don't forget to have fun. This is supposed to be enjoyable. If you're not having fun, you're doing it wrong.
The machine is a mirror. The images you create are a reflection of your own creativity, your own biases, and your own imagination.
This is the most creative tool I've ever used. It has completely changed the way I think about design, art, and creativity itself.
We are just at the beginning. This is the first inning of a nine-inning game. The best is yet to come.

I know 99 lessons is a lot to take in. Don't feel like you need to memorize them all. Just pick a few that resonate with you and start there. The most important thing is to just start. Start creating, start experimenting, and start learning. The future of design is here, and it's more exciting than I ever could have imagined. Now go build something amazing.

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.

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.

How were these items selected?

Each item on this list comes from direct experience, either from building my own companies or from patterns I've observed across the 200+ startups I've invested in. I prioritize practical, actionable items over theoretical concepts.