If you’ve been snooping around the darker corners of the internet recently, you might’ve noticed an odd new trend among the meme community. Depictions of strange, blurry figures have been lighting up Reddit with captions like “Winnie the Pooh Chicago Drill” or “Bugs Bunny signing the declaration of independence.”
If these photos seem unnatural, they are: An artificial intelligence (AI) created them. Humans enter a few words and then the AI turns them into images – with some disturbing results. But why the sudden spotlight on AI art?
You might argue it’s the result of the super AI art program, DALL-E by San Francisco-based, OpenAI in 2021. Along with the launch, they open-sourced their code spawning a horde of other AIs. Whatever the reason, Angus Russel isn’t complaining: He’s been trying to generate buzz around his AI-art generator, NightCafe Studio, for four years.
It began as a program to help people create more personalized wall art, but Angus found people spent most of their time making art, not printing it. After losing money for a year running servers, he finally limited the number of free creations people could make on his website. Suddenly, his inbox flooded with requests asking how they could buy more creations.
Today, Angus’s website makes roughly $100,000 in MRR and customers create as many as 120,000 pictures on his site per day. He’s no longer the “the guy who needs to buy some art” but “the guy who runs that cool art business online.”
You Need Some Art in Your Apartment
For someone who now makes his living helping people create digital art online, Angus was not exactly a connoisseur of the craft four years ago.
“I had a guest over to my house in Sydney in 2018,” he says. “They commented that I needed some art because my walls were ‘very bare’.”
Angus started browsing online for pictures he could hang on his walls to spruce the place up, but he couldn’t find anything he liked.
“I’m not usually a compulsive or obsessive person, but I looked through hundreds of pages of artwork and didn’t see anything good. I wanted something a bit more personal. Something I could tell a story about.”
During his search, Angus stumbled into the growing realm of AI-generated art. At the time, most of it involved the “neural transfer” technique, where an AI recreated a photo in a given art style.
“I thought that someone out there was using these AI for personalized prints, but I had a search and couldn’t find anyone.”
He found a simple neural transfer AI online, created a website that would run images through it, and attached it to a store on Shopify that let people order prints. He immediately sold some orders and knew he was onto something – he just wasn’t sure what.
People Don’t Want to Buy Prints
Although Angus had sold some art and decorated his walls, his AI-generated prints were putting him in the red financially. “Most people would create art but wouldn’t buy. It cost me money because the AIs need a lot of GPU power. Generally, I lost five cents per piece of art generated – fortunately, it wasn’t very popular.”
Angus absorbed the losses and continued developing the AI, hoping for later profits. He almost sold the business for a small sum of $5,000 but thought better of it. Then his fortunes changed when a customer cost him a lot more money than he was comfortable losing.
You Don’t Need to Sell a Physical Product
One day, Angus checked his server costs and his jaw hit the floor. “One person had created 1,400 pieces of art in a week,” he says. “That cost me a few hundred dollars – so I limited the number of creations people could make. I created a credit system for seven artworks per person and I made it impossible to get more credits.”
The same night he deployed the credit limit update, he received six emails from fans of his site asking how to get more credits so they could make more art. He created a payment tool to unlock extra credits and immediately became profitable. Then in 2021, OpenAI came out with the original DALL-E AI art program1.
Angus credits DALL-E with showing people the power of AI in art. DALL-E also included a new tool: text-to-art. People could type a description of the image they wanted to create and let the AI guess how it should look. Often, the resulting image was quite convincing.
On the heels of DALL-E, a group of netizens created a free AI text-to-art generator. Angus added the feature to NightCafe Studio, and a couple of months later, his platform went viral.
“Someone on Reddit posted an image they did on our platform to a big SubReddit. The post blew up and although he didn’t mention us, someone posted that it was from our site in the comments. I woke up the next morning and NightCafe Studio wasn’t working anymore – we’d been hit by the Reddit hug of death. The day that happened, we did seventeen times more sales than the day before.”
Angus struck a nerve on the internet by allowing people to become artists with a few keystrokes. “Text-to-image is very addictive. It gives you the feeling you get from accomplishing something very easily. That’s what keeps people coming back.”
Today, Angus says his business attracts customers of all types. He’s still trying to figure out his core audience. “It’s hard to define a target market. Most people come for the community and that feeling of being creative. For many, it’s a therapeutic activity as they’re winding down before bed. Some people take a scientific approach to figuring out what our AI is good at. Others are just using it to create NFTs. Very few use it for its original intent of printing images.”
It’s Hard to See AI Getting Much Better
Competition soon began nipping on the heels of Angus’s success. Some have copied large parts of his model, he says. He hired a small team including a developer to update the website and a social media manager to increase profits. The team created a reasonably popular Instagram account (@nightcafestudio) and other assets to increase traffic to the site. Today, he says keeping his team lean has helped him stay ahead of the pack.
“We’re still the leader in many ways,” he says, “but the space is heating up quickly. One competitor has a lot more users, but is completely free and probably not profitable yet. Another competitor has a better algorithm right now but is currently in a closed beta. DALL-E 2 will open up to everyone at some point too. There are also open-source efforts to replicate DALL-E 2 and Google’s text-to-image algorithm, Imagen.”
Interestingly, Angus thinks text-to-art AI may not have much further to go.
“We have this thesis that AI art will begin to taper. There’s not much room for major improvements. I don’t think there will be many more improvements beyond DALL-E 2. If you follow the OpenAI Instagram account, it’s hard to believe another program could do better. Future innovations in the field will likely be in commodification and pricing, not AI ability.”
Angus credits his success to three things: being lucky, persevering in the face of lost capital, and listening to his customers.
We figured out that building a good community and good UX was more important than specializing in the technology.
“All of our features are based on user feedback. We figured out what brought people back and leveraged that. Ironically, we’re a company working with AI without an AI specialist on staff. We figured out that building a good community and good UX was more important than specializing in the technology.”
Angus’s story proves that you don’t need to be an expert to build a successful startup. Angus was neither an art aficionado nor an AI specialist. Instead, he focused on what he was good at: user experience. As a result, he created a friendlier product that was more accessible to the average audience.
1https://www.theverge.com/23162454/openai-DALL-E-image-generation-tool-creative-revolution
Want to share your bootstrapping story with the world? Enter your details in this form and we’ll be in touch.
