Midjourney: The Billion-Dollar Company That Lived in a Chat App

August 2022, Pueblo, Colorado. A man submits a painting to the digital arts competition at the Colorado State Fair. The title? Space Opera Theatre. It is stunning. A baroque hall, flooded with golden light, a massive circular window looking out into the cosmos, figures in robes staring into the void. It looks like a Renaissance painting of Star Wars.

The judges pin the blue ribbon on it. First place!

The artist, Jason Allen, accepts the prize. And then, he tells them the truth. He didn’t paint it. He didn’t photograph it. He didn’t even use Photoshop.

He just typed some words into a chat room.

The news broke the Internet. Twitter threads proclaimed the death of art. Artists called it theft. The judges admitted they didn’t know what was going on.

But here is the paradox ☝️.

At that exact moment, the company behind that image, Midjourney, didn’t have a marketing department. They didn’t have venture capital. They didn’t even have a website where you could log in and generate an image. To use the product, you had to join Discord, a chat app for gamers.

Yet within three years, Midjourney would have over twenty million registered users and nearly five hundred millions of annual revenue. No investors. A team of just 40 people. And lawsuits from Disney, Universal, and DreamWorks.

So how did Midjourney, a product that lived entirely inside a chat app for gamers, grow so big? How did a product with such a terrible user interface get so many users? How did a company with no traditional app and no venture capital manage to beat juggernauts like OpenAI and Google? And how did they find themselves fighting lawsuits against Hollywood’s most powerful studios?

This is the story of Midjourney, the leanest high-growth AI company in the world.

THE SEED

David Holz is not your typical Silicon Valley CEO.

If you met him in the street, you might not think business mogul. You might think he looks like the guy who fixes your Mac at the genius bar.

Before Midjourney, Holz was a student of the universe, literally. He studied applied math and physics. He worked at the Max Planck Institute. He did time at NASA’s Langley Research Center, working on LiDAR technology to map Mars.

Then he left in 2010 to found a startup called Leap Motion. They were building a little device that let you control your computer with hand gestures. Like Minority Report. It was incredibly hyped. It raised millions of dollars and it worked… well… technically.

In practice, in the end, it failed. The market adoption was just underwhelming, and Leap Motion sold to the British company Ultrahaptics in 2019 for roughly $30 million dollars, about 10% of Leap Motion’s $300 million peak valuation in 2013.

Why? Because it was a solution looking for a problem. It was cool technology, but it didn’t really have a Job to Be Done. Waving your arms at a screen is tiring and makes you look kinda silly. Turns out, the mouse is pretty great!

Holz learned the hard way: impressive tech is worthless if it doesn’t solve real problems. And distribution often matters more than innovation.

So he retreated. He started thinking about new ambitious ideas in lowkey mode, like a lion waiting for a prey to pass nearby.

In 2021, the prey manifested.

There was a buzz around new AI research papers on something called diffusion models. Holz was obsessively reading the diffusion model papers. These new AI systems could generate images from text.

The results were messy, unpredictable. But Holz saw something others didn’t. He wondered: “What does it mean when computers are better at visual imagination than 99 percent of humans?”

He saw a new medium of thought emerging. A way to democratize visual arts. A way to empower billions of people to draw and paint anything they could imagine without having to sweat ten years learning.

So in 2022, he founded an independent research lab. They called it Midjourney. No external investors. Just 11 engineers and a Discord server. With a big hairy goal: expand the imaginative powers of the human species.

Sounds pretentious, right?

Well… it worked.

THE SPARK

In late 2021, they had a prototype, but the initial tests were… unexciting.

You’d give someone this magical canvas and ask, “What do you want to see?” And the most common answer would be a disappointing, one-word: “dog.” Maybe two words, like “pink dog”. Or three words, like “pink fluffy dog”.

But at some point, they made a breakthrough. It was not a technical breakthrough, it was a social breakthrough.

They noticed that when people were in a group, the dynamic changed. One person would say “dog”. Another would say “space dog”. A third would shout “Aztec space dog!”.

At that moment, Holz got it. They were not building a tool. They were building a social network. So they would not build a standalone app. They would build a bot for Discord.

On paper, this looked like a dumb idea. Why would you start with a Discord bot instead of a normal website everyone already knows how to use?

When Midjourney launched in March 2022, it should have flopped.

They had this weird Discord chat bot social network idea. And OpenAI’s DALL-E already dominated the conversation. DALL-E 2 was the polished, invite-only frontrunner. Stable Diffusion was coming. Google had Imagen in the wings. These were billion-dollar research labs with massive datasets and marketing machines.

Midjourney had… Discord.

Instead of building a normal website or app, Holz forced every user into Discord. Textbook marketing logic would say this is suicide. You are adding massive friction to the product. You require users to download a third-party chat app, join something called a server, and type obscure commands, like slash imagine, in a public channel. And then, people would often lose their images in that stream of thousands of other people’s messages posted on that public channel.

But let’s look closer.

By forcing users into public channels, Midjourney solved the cold start, chicken-and-egg problem every social network faces. When you joined, you didn’t see a blank text box. You saw a waterfall of other people’s artworks. You saw the prompts they used. You saw the results.

You learned how to use the product by watching others. You saw someone generate something incredible and thought, “I want to do that”, and you could then just copy-paste the prompt they used.

Every user was generating content that advertised the product to every other user. They turned a single-player utility into a massively multiplayer game.

Holz said in an interview with The Verge, quote: “We could have built our own social network, but why? Discord already has the social primitives. The community is already there.” End quote.

In another interview, he told Ben Thompson from Stratechery, quote: “I think that the Midjourney experience would not work at all if it was just talking to a chatbot in a room by yourself, but the second that it’s in a room with lots of people, it becomes really interesting.” End quote.

But beyond this social aspect, Midjourney had another advantage. One nobody expected would matter so much.

Beauty.

Midjourney launched with a radically different model tuning. It didn’t try to look like a photograph; it tried to look like concept art. It understood lighting, composition, and texture. It was “opinionated” software.

While other models were optimizing for accuracy, Midjourney optimized for aesthetics. They realized that 99% of users are not artists.

When normal people type something and get a bland drawing, they get bored, and leave. If they get a cinematic, moody masterpiece, they feel like a genius, and stay.

Midjourney images felt designed, intentional, cinematic. The output wasn’t just correct. It was beautiful. Midjourney made its users feel badass!

As 37Signals CEO Jason Fried once said: “Here’s what our product can do” and “Here’s what you can do with our product” sound similar, but they are completely different approaches.

The real goal is to make your users feel badass!

Experimental artists. Creative geeks. Hobbyists who cared about mood and atmosphere. These were the early adopters. And they were ecstatic. They were vocal.

By July 2022, just six months after launch, the Discord server reached one million users, surpassing major servers like Fortnite and Minecraft at the time. Press coverage in The Verge, Forbes, PCWorld. The Economist even put a Midjourney-generated image on its cover. Holz described this growing community as a “hive mind of people, super-powered with technology”.

For any normal startup, that would be the moment of raising more capital to accelerate growth. Going from seed funding to series A funding round. Or from Series A to Series B and so forth.

But that was not how Midjourney operated. They were not burning loads of VC cash like Silicon Valley cool kids. They were charging from the start.

Subscription tiers. Pay for compute. A business model aligned with usage, not vanity metrics. And they stayed lean. Just a few dozen people at that time. No bloat.

In August 2022, Holz said in an interview with The Register, quote: “We’re like a self-funded research lab. We can lose some amount of money. We don’t have like $100 million of somebody else’s money to lose. To be honest, we’re already profitable, and we’re fine”. End quote.

They were going to ride this momentum to mainstream status, on their terms.

GOING MAINSTREAM

Geoffrey Moore’s “Crossing the Chasm” theory says there is a gap between early adopters, i.e the tech nerds, and the early majority, i.e the mainstream.

The early adopters loved Midjourney. But early adopters always love new things.

The chasm, the gap between enthusiasts and normal people, that’s actually where many products fall and die. Early adopters want possibilities. Early adopters tolerate quirks and friction.

The mainstream wants reliability. The mainstream hate bugs.

Midjourney crossed this chasm with three big things.

First big thing: quality improvements. In November 2022, they released Version 4. The leap was visible. Images went from “impressive for AI” to “wait, a human didn’t make this?”.

Lighting improved. Details sharpened. Consistency increased. The model started feeling less like a slot machine and more like a reliable collaborator.

Then Version 5 in March 2023. It’s even better. Photorealism. Greater consistency. The “shimmer” and weird artifacts of early AI are gone. Fingers start looking like fingers. Eyes look like eyes. The kind of output you could use commercially. For marketing. For book covers. For social media. For everything.

Second big thing: the memes. This is where Midjourney got out of tech circles and broke into mainstream conversations.

March 2023. Someone generates an image of Donald Trump being arrested. Surrounded by police. Dramatic lighting, facial expressions, and dynamism. It goes viral. Millions of views. News outlets cover it. Some people believe it’s real, until they figure it out. The conversation shifts from “Is this real?” to “Holy shit, AI can do this?”

A week later, we get an image of Pope Francis wearing a massive, stylish white Balenciaga puffer coat. It looked swaggy. It looked real. It also goes viral instantly. Fashion blogs cover it. People think the Pope just got a fashion epiphany!

These viral images proved Midjourney could create things people cared about sharing. Things that sparked emotion. Delight. Confusion. Outrage. It didn’t matter which. What mattered was they couldn’t be ignored. It was a public demonstration that this tool could generate impactful content in seconds. It was a tipping point in public awareness.

Third big thing: the web interface. For two years, Midjourney lived only in Discord.

That was brilliant for the early community of early adopters. But it was a total pain for everyone else. Non-gamers didn’t use Discord. Professionals found it clunky. The mainstream needed easier.

So in 2024, Midjourney publicly launched their web editor.

Simple interface. Gallery of examples. Create without joining a server. The barrier to entry disappeared.

Midjourney recognized that to reach the Late Majority, they eventually had to leave the chaos of Discord. But they only did it AFTER the community was cemented.

Midjourney was now not just for tech-savvy creatives. It was for marketing teams. Publishers. Social media managers. Real work.

The numbers speak for themselves. From 1 million Discord members in mid-2022 to over 20 million by mid-2025. Daily active users soared into the millions. Revenue, without a cent of VC money, exploded from an estimated $50 million in 2022 to $500 million in 2025.

In 2024, Stripe even named Midjourney its most globally distributed merchant and gave David Holz a custom CRT monitor displaying a live map of the company’s worldwide customer base.

They had built a new market, and were leading it. They’d crossed the chasm.

But here’s the thing about crossing the chasm. A smooth landing is absolutely not guaranteed.

HEADWINDS

The first challenge to hit Holz and the Midjourney team was that competition got very heated, very fast. By 2023 and 2024, the moat began to dry up. Stable Diffusion offered open-source control. Adobe integrated Firefly directly into Photoshop. Making pretty pictures was becoming a commodity.

Then, OpenAI embedded DALL-E into ChatGPT. ChatGPT’s massive user base now had image generation built in. That integration went viral and people suddenly started sharing Studio Ghibli versions of EVERYTHING.

And video arrived. Kling. Runway. Veo from Google. Grok Imagine from xAI.

Midjourney had carved out a new market. But that market was getting crowded.

Unfortunately, external competition wasn’t Midjourney’s only problem. They were also fighting battles from within.

The second challenge was moderation, which became a nightmare. Remember, Midjourney’s entire model relied on public generation. But public generation also means public abuse.

In 2024, ahead of the U.S elections, Midjourney started blocking images of certain political figures. Xi Jinping. Donald Trump.

The goal was to prevent deepfakes and misinformation. But the backlash was immediate. Users accused them of censorship. Of choosing sides. Of overreach.

Holz defended it as risk management. But the tension was real: how do you run a platform for imagination when imagination can become a liability?

Third challenge, and this is the big one: copyright.

In 2023, a Twitter user posted a children’s book they’d created entirely with Midjourney. Illustrations, characters, everything. Twitter got very mad.

Artists pointed out the images looked suspiciously similar to existing illustrators’ styles. The question wasn’t subtle: did Midjourney train on copyrighted work without permission? The guy took a proper Twitter beating.

Then the lawsuits started.

A class-action lawsuit filed by artists Sarah Andersen, Kelly McKernan, Karla Ortiz, alleged that AI companies like Stability AI and Midjourney scrapped billions of copyrighted images to train its model without consent. The allegations were specific: training datasets included copyrighted art from living artists, used to generate images that directly competed with those same artists.

Then in June 2025, the hammer dropped.

Disney, Universal, and DreamWorks filed a federal lawsuit.

The claim: Midjourney enabled users to generate images of Mickey Mouse, Shrek, characters from Frozen, copyrighted IP worth billions. The studios alleged the model was trained on their work and now functioned as an unlicensed content generator.

Warner Bros Discovery followed months later. Their claim was devastating: Midjourney had built a business on the “theft” of their iconic characters, from Superman to Scooby-Doo.

This was not just about artists anymore. This was legal warfare from industry giants. That kind of giant with bottomless pockets. That kind of lawfare that could shut down a company.

But challenges were also cultural. AI-generated images started flooding the internet.

Social media feeds began filling up with low-effort AI garbage. “Shrimp Jesus.” Fake architectural wonders.

More recently, OpenAI released SORA 2, an AI short video generator for social media which accelerated the trend. At some point, the novelty wore off. People started calling it “AI Slop”. And Midjourney, because it was so visible, so successful, became one of the targets of that backlash.

The narrative shifted from “democratizing creativity” to “stealing from artists.” From “collaborative imagination” to “copyright infringement.” From “community-funded underdog” to “capitalist thief.”

So what did Midjourney do?

They fought. They hired lawyers. They filed motions. They argued fair use and transformative purpose. They pointed out that users controlled prompts. Midjourney didn’t create copyrighted images, users did.

Midjourney didn’t apologize, but they adapted.

They didn’t pivot to B2B enterprise sales to hide. They doubled down on their power users. They launched video generation mid-2025, competing with Veo and all the others. They released “Style Tuners” to give artists more control. They improved moderation tools. They added features for commercial licensing and attribution. They evolved their product. They stayed small. They stayed quiet in the press. They kept shipping models. V5. V6. V7.

And they invested some more in the community.

Because here’s the truth: Midjourney’s strength was never just the model. It was the people. The Discord channels where users taught each other prompt craft. The galleries where people showcased their work. The rituals of iteration and upscaling and sharing.

That community didn’t abandon them. Even as lawsuits mounted. Even as critics called them thieves. The users stayed. Because Midjourney gave them something they couldn’t get anywhere else.

A fellowship of people making beautiful art with just a keyboard and a screen.

CONCLUSION

So, where does that leave us?

Midjourney started as a Discord bot run by a guy who used to map Mars. It became the defining visual engine of the early 2020s. They broke major rules in the marketing playbook. They didn’t build an easy-to-use web application. They didn’t take venture capital. They charged real money from the start instead of artificially boosting growth with free users. And they still achieved incredible success.

Now that competition in image and video generation is fiercer than ever, what will Midjourney do in the months and years to come? Keeping up with the competition for sure, but probably, not just that.

Just like Space Opera Theatre shocked the world in 2022, Midjourney may be preparing its next surprise.

On August 28, 2024, Midjourney publicly announced on Twitter they had started working on hardware.

They had hired Ahmad Abbas just a few months before. Abbas, who is an ex-Neuralink engineer and the former hardware engineering manager on Apple’s Vision Pro, is also a former colleague of Holz at Leap Motion.

For now, nothing is known about the hardware they plan to make: AI-generated 3D worlds? Real-time generated video games? Something even crazier?

When Holz revealed the initiative to Midjourney users on Discord in early January 2024, he called it a device for creating and managing thousands of 3D rooms. He later alluded on X that Midjourney “will make the Orb.”

So what’s the orb? I don’t know, and maybe even Holz doesn’t know either.

Yet.