Picture this: you’ve finished your fantasy novel. The manuscript is polished, the beta readers are happy, the formatting is done. Now you need a cover. A good one. The kind that makes someone stop scrolling on Amazon and think, “okay, what’s this about?”
You could hire an artist (expensive, slow, and you might not love the result). You could use a stock photo (and end up with the same shirtless man holding a sword that graces forty other books in your genre). Or you could type a description of what you see in your mind and watch an AI paint it for you in under a minute.
That last option is Midjourney. And the reason it’s worth talking about, three years into the AI image revolution, is that it doesn’t generate images the way most tools do. It paints them. The difference is hard to explain until you see it, and then it’s hard to unsee.
The Guy Who Turned Down Apple Twice
David Holz’s path to building Midjourney makes more sense once you know he grew up with a father who ran a dental practice on a sailboat, sailing around the Caribbean to reach patients. Unconventional thinking, apparently, runs in the family.
Holz studied math and physics at UNC Chapel Hill, worked as a contractor for NASA, did neuroscience research at the Max Planck Institute, and then dropped out of his PhD to co-found Leap Motion, a gesture-recognition hardware company. Leap Motion let you control computers with hand movements, like something out of Minority Report. Apple wanted to buy it. Holz said no. Apple came back. He said no again. The company eventually sold to another firm for about $30 million, and Holz moved on.
What he moved on to was a question that had been nagging him: what if computers could help people imagine things? Not just execute commands or organize data, but actually expand what a human mind could visualize?
In August 2021, with a team of ten engineers and zero venture capital, he founded Midjourney. The first public version launched as a Discord bot in early 2022. By August of that year, six months in, the company was already profitable. No outside investors. No marketing budget. No PR firm. Just a research lab in San Francisco that spread by word of mouth because the images it produced made people stop and stare.
Today, Midjourney generates roughly $500 million in annual revenue with around 100 employees. Holz still makes announcements the same way he always has: by posting messages in Discord.
What Midjourney Actually Does
At its core, Midjourney is simple. You describe an image in words, and the AI creates it. A dark forest with bioluminescent mushrooms. A weathered lighthouse on a Scottish cliff at dawn. A woman in 1920s Shanghai stepping out of a taxi in the rain.
The interface has evolved. Originally, you had to use Discord, typing commands like /imagine into a chat channel alongside thousands of other users creating images simultaneously. That communal chaos was actually part of the design (more on that in a moment), but it was also a barrier for people who just wanted to make pictures without learning how Discord works.
In 2024, Midjourney launched a full web editor at midjourney.com. You can now prompt, edit, inpaint (change parts of an image while keeping the rest), extend canvases, and organize your work in a visual interface that doesn’t require knowing what a Discord server is. For authors who want to iterate on a book cover concept without navigating chat channels, this was a major quality-of-life upgrade.
The current model, Version 7, also introduced voice prompting, which lets you describe what you want out loud. It’s a surprisingly natural way to work when you’re still figuring out the mood of a scene and don’t want to fuss with exact wording yet.
Why Authors Keep Coming Back to This One
There are dozens of AI image generators now. DALL-E, Adobe Firefly, Leonardo, Ideogram, and more. Midjourney’s staying power comes down to one thing: aesthetic sensibility.
Most AI image tools optimize for accuracy. You describe a red ball on a table, you get a red ball on a table. Midjourney optimizes for something harder to quantify. It has a sense of composition, light, mood, and style that makes its output look less like “AI art” and more like concept art from someone who went to art school.
For book covers, this matters enormously. A fantasy cover needs atmosphere. A literary fiction cover needs restraint. A thriller cover needs tension in the lighting. Midjourney handles these tonal demands better than most of its competitors because it wasn’t trained to be literal. It was trained to be evocative.
Version 7 added a personalization feature that takes this further. You spend about five minutes rating a set of images (which do you like better, this one or that one?), and Midjourney learns your aesthetic preferences. From that point forward, your generations subtly reflect your taste. Two authors typing the same prompt will get different results, tuned to what each one finds visually appealing.
This is Midjourney’s Purple Cow, if you will. It’s not trying to be the most obedient image generator. It’s trying to be the most tasteful one.
The Discord Experiment
Holz made a deliberate choice to launch Midjourney inside Discord, and his reasoning was more interesting than “it was easy to build a bot there.”
He wanted image generation to be social. His argument: making images alone, talking to a chatbot, is fine but limited. Making images in a room full of other people making images turns the process into something closer to wandering through a gallery. You see what other people are creating. You get inspired by prompts you never would have thought of. You borrow ideas, remix styles, and develop your visual vocabulary by watching the community work.
The Midjourney Discord server grew to over 21 million members, making it the largest community on the entire platform. For many users, it was their first real interaction with AI-generated images, and the communal experience made it less intimidating than sitting alone with a blank prompt box.
The web app now serves users who prefer a quieter, more private workflow (the Pro and Mega plans include a Stealth Mode that keeps your images out of the public gallery). But the Discord community remains active and is still one of the best places to learn what’s possible.
What You Should Know Before Subscribing
Midjourney is impressive, but it has real limitations, and being honest about them will save you frustration.
Text in images is a mess. This is the single biggest limitation for authors making book covers. Midjourney scores poorly at rendering readable text. If you need a title on your cover (and you do), plan to add it yourself in Canva, Photoshop, or another design tool. Never rely on Midjourney to spell your book title correctly. It probably won’t.
Characters drift across generations. Version 7 introduced character reference tools that improve consistency, but if you need the exact same character looking exactly the same across twenty illustrations (say, for a children’s book), the character will start shifting after a handful of images. Midjourney works best for one-off character portraits or scenes where slight variation is acceptable.
The learning curve is real. Expect to spend 10 to 15 hours before you’re consistently producing results you’re happy with. Midjourney rewards experimentation and prompt-crafting in ways that aren’t immediately intuitive. Parameters like --ar (aspect ratio), --stylize (how much artistic liberty the AI takes), and --chaos (how much variation between results) all affect output, and learning to use them well takes practice.
Precision is not the strength. If you need exactly five birds in a sky, or a character holding a sword in their left hand specifically, Midjourney will often interpret rather than obey. It’s brilliant at mood and composition but unreliable for technical accuracy.
There’s no free tier. Midjourney briefly offered free generations at launch but discontinued them. The Basic plan starts at $10/month. Annual billing saves 20% across all tiers.
Copyright questions are unresolved. Several lawsuits involving Midjourney’s training data are ongoing. For authors using generated images commercially, this legal landscape is worth monitoring, though it hasn’t stopped hundreds of thousands of creators from using the tool.
Who This Is For (and Who It Isn’t)
Midjourney is a strong fit if you’re an author who needs visual assets and you value atmosphere over precision. Book covers for fantasy, science fiction, literary fiction, and horror genres benefit most from Midjourney’s artistic strengths. It’s also valuable for worldbuilding visualization, character reference art, social media graphics, and the kind of mood-setting imagery that helps you stay immersed in your story while you write.
It’s less ideal if you write in genres that rely on photorealistic cover styles (many romance and thriller subgenres, for example), if you need perfectly consistent character illustration across a series of images, or if you need text rendered directly into the image.
And if you’re someone who wants to type a prompt and get exactly what you described with no interpretation, Midjourney will frustrate you. It’s an artistic collaborator, not an instruction-follower. You guide it, and it surprises you, and often the surprise is better than what you originally had in mind. But “often” is not “always,” and you’ll burn some GPU hours learning the difference.
The Bottom Line
Three years in, with hundreds of competitors, Midjourney is still the tool that makes people stop and say, “wait, AI made that?” It earns that reaction not through photorealism or technical accuracy, but through something harder to engineer: taste.
David Holz built a self-funded research lab that turns a profit, takes no outside money, does no marketing, and grew to half a billion dollars in revenue on the strength of its output alone. That tells you something about the product.
For authors, the practical case is straightforward. If you need visual assets for your books and you’re willing to invest the time to learn the tool, Midjourney will produce artwork that looks like you hired an illustrator, not like you asked a computer to draw something. You’ll still need to add your own text. You’ll still burn through some failed experiments. But the images that land, and they do land, will be the ones that make readers pause on your book cover and wonder who your artist is.
You don’t have to tell them.
