From conception to validation — How I came up with a side project idea, and sold 20+ AI Generated Art pieces to validate it.

Mid last year, my girlfriend and I had a house guest who commented that our walls looked a bit bare. It’s not something we’d really noticed before, but once it was mentioned, we realised they were right — we needed some art. I thought finding some interesting art online would be a pretty quick, easy process, but I soon found that the sheer volume of available art turned me into a critic and I couldn’t find a piece that had meaning to me.

I wanted art that reflected something about me — even if that just meant a landmark from my city, or a subject based on one of my hobbies or interests. After scrolling through hundreds of pages of artworks on every wall-art store I could find, I was still left unsatisfied.

Then I had an idea — one of my interests is technology, and in particular AI. What if I could buy some AI-generated art? Wouldn’t that be a conversation starter! Even better, what if I could create some AI generated art myself? It could be ultra personalised — based on a photo from one of my holidays, or a picture of my dog, or a landmark from my city. Why didn’t I think of this sooner!? AI generated art has been around for years now, there must be plenty of places I can generate and buy AI art.

So I Google “Buy AI Generated Art”. Hmm, I get one shop with a handful of pre-generated pieces, a heap of articles about AI and art in general, the guys who sold that AI generated portrait for $432k (a bit out of my price range), and Deepart.io — which is the closest thing to what I was imagining, but buying prints seems like an afterthought. Further searches didn’t reveal any better options (although there are other art generators out there, none offer much in the way of buying your artwork as a poster or canvas print).

To this point I hadn’t been considering building my own AI Art app. I thought for sure it must already exist. But when my continued Googling didn’t yield what I was looking for, the wheels started turning. I jumped on Slack and messaged some friends who are also into startups and side projects. The idea started to excite me.

I’ve been working on side projects and startups my entire career, so I knew the first step wasn’t to build the app (a mistake I’ve made before — more than once), I first had to validate the concept. Would other people want to decorate their walls with AI-generated art, or was it just me? Would they pay? How much? These questions all needed answers, ideally before writing a single line of code. So I started thinking about how I could answer these questions as quickly and confidently as possible.

What I came up with was simple! All I had to do was find an open-source Neural Style Transfer algorithm on Github, generate some interesting art with it, throw it up on a Shopify store and point some Facebook ads at it. If people buy AI art that they didn’t even create themselves, surely they’d buy their own personalised creations.

I was too busy to do it right then, so I decided to sit on it for a while. I had to finish a contract job, then I had a holiday booked where I planned to propose to my girlfriend. I figured if I was still excited about the idea after all that, I’d throw the MVP together to validate the idea. If it showed some promise (I.e. if people actually bought some art), I’d think about building out my vision — an AI art generator focused on creating personalised art to decorate your house with.

A few months went by. I finished my contract, got engaged (!!), added some features to another project (shoutout: BrainBlocks) and suddenly I had some free time again. I hadn’t forgotten about AI Art. I decided to go for it.

I found some Neural Style Transfer libraries on Github and tried to get them running. I don’t really know Python, so it was a bit of a challenge, but I eventually managed to run one on my Macbook. It took about 50 minutes to run and the result didn’t look that great. I needed to be able to run them much faster to be productive. A bit of Googling led me to Google Colab — a Jupyter Notebook in the cloud that runs on GPUs. The library I’d chosen was written for use from the command line, so I needed to do some refactoring to get it running in Google Colab. Thankfully Python’s syntax is pretty easy to learn, so I managed to get it working pretty quickly. On Google Colab it ran in 5–7 minutes, which was still pretty annoying, so I built a queue system where I could drop the inputs and styles in Google Drive, and define the jobs as rows in Google Sheets. That let me queue as many jobs as I wanted and leave them to run. It might sound impressive, but Google Colab made it really easy. With this system I quickly got a feel for what would work well and what wouldn’t, and was able to make some interesting artworks like the Basquiat Pineapple, Starry Night Astronaut, and Majestic Fox.

Some pieces from my first round of artworks.

Now that I had some artworks to sell, I needed two things: a shop to sell them in, and a printer to print them. I found a printer that would suit, ordered some samples and was happy with the results, so the next step was to build a store. I knew from the start I was just going to use Shopify. I threw the products up on the default Shopify theme, added a few essential plugins, spent about an hour coming up with a business name that had a domain name available, landed on NightCafe.studio, and I was ready to sell.

One of the original samples, hanging in our lounge room in Sydney.

Because I’d visited every poster and canvas art store on the internet, Facebook had been serving me ads for artworks for months. I’d picked up on some of the tactics they used and noticed that I’d been flicking through a lot of the carousel ads to see if there was anything good, so I decided to make a carousel ad on Facebook to bring some traffic to the site and hopefully make some sales. I had to learn the Facebook ads manager and read about how to make good ads, but it didn’t take long before I was serving carousel ads to people who were interested in art and artificial intelligence.

It took almost a week — and a few hundred dollars of ad budget — to make the first sale. The sale didn’t come close to paying for the ads, but it proved that people (well, at least one person) were willing to pay for AI generated art. Optimisation could come later.

I spent about a month trying to optimise the website and Facebook ads. The most important update I made to the website was an exit-intent popup (those annoying popups that come up when you move your mouse towards the close-tab button) that surveyed the user about why they were leaving without buying anything. I usually hate those things, but they have their place, and I learned a lot about what people wanted through that survey.

I also quickly learned that I was doing Facebook ads totally wrong — I should have made 20 ads and given them a budget of $5/day each, then kill the ones that didn’t perform and slowly ramp up the ones that did. What I had been doing was running one ad with my entire budget. I started again from scratch and had a bit more success, making about 20 sales over the next two months. The Facebook ads still cost a lot more than the sales I made from them, but by this time I was convinced that there was a market for my vision — an AI Art generator oriented around creating art for your home.

In the past I would have leapt into building something like this before doing any kind of idea validation. I’ve always known you need to validate your ideas before diving in, but it’s so easy to make excuses, or trick yourself into thinking you’ve done enough validation because you just want to start building. It’s taken a lot of side projects — and many failures — to really learn the importance of validating an idea before committing to building a product that no-one really wants. I’m happy with the process I followed this time. I’ve proven that people are willing to buy, which means the challenge now is to build a great product, and get it in front of the right people.

I’m planning a whole separate post about how I built the creator itself, so right here I’m going to skip ahead and point you to NightCafe Creator — my AI Art generator app. There’s still a lot I want to add to it and improve, but it works. It has a clean interface, it’s installable as a mobile or desktop app (It’s a Progressive Web App), and even has some unique features like multiple styles and style masks.

I’d love it if you gave it a go. Create something interesting and post a link to it in the comments! And if you have any feedback (good or bad) I’d love to hear that too.

Keep an eye out for my next post, which will be about how I built the actual app (that’s the stuff I’m actually good at — unlike Python and Facebook marketing). If you want to stay up to date with my journey, and news about NightCafe, please subscribe to the newsletter and follow me on Medium, Twitter and Reddit.

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store