As a developer familiar with AI and NFTs I was curious how long it would take me to create an NFT project and sell out from absolutely zero.
Here’s a documentation of my journey.
I created a channel in a non-crypto related Discord server that I frequent to serve as my dairy. In here I would record updates, thoughts and progress.
When recording time I was recording my real world actual time spent, not the wall time. I knew I was going to run an AI to generate art, I would be away from the computer for a lot of the time while it did all the hard work.
Knowing that there is a large set of AI image generation tools that rely on inputs I decided to enlist some help from my friends.
I created another channel in the Discord for them to post words. The channel was called #word-collection. I gave a brief summary of the purpose and then openned it up for people to submit words that are meaningful to them.
It went about as well as you can imagine.
I left the chat to see what would happen.
There are hundreds of AI image generators online. Art Flow was one that was quite promising. I tested out a couple but quickly decided the quality was not that good and the time taken to produce an image was too slow.
I did a quick cost analysis. 0.2 ETH for contract deployment. IPFS price is trivial. So is website domain. Heroku free tier would work for hosting the front end.
I decided on a stealth launch and free mint because I knew that would sell out quickly. I estimated that at a 5% royalty on secondary sales, I would be able to make back the initial investment in no time.
Total time spent so far: 30 mins
I’ve played with NFT contracts before so a lot of this was copy/paste from other projects. I added in a Series feature, planning that if this launch was successful I could further enhance the collection by releasing additional Series of content.
That never happened.
Not a lot else to say here. Including unit tests and deployment to Rinkeby, I spent about 1 hour at this point.
I stumbled across Text2Art, an open source (MIT licensed) text to image tool that I could download and run locally. I set it up to run for 100 iterations and then went to sleep.
When I woke up, it didn’t stop. 9.5 hours of generating later and I get the first image I would include in the collection. The “Smooth Shark”.
You can now find the shark on Opensea. It is one of my favourites.
The generator also output a video of the generation process, which I have included in the NFT.
After testing a couple more prompts like Orange and Perspicacity, I figured this was the best way to go for images.
I spent a long time playing with the image generator, altering the code to collect only the correctly spelt nouns from the Discord channel.
I have now opened the generator repository for anyone to use. Please respect the associated licenses.
I uploaded some of the metadata to IPFS to make sure it worked in Opensea.
One thing I learnt here was that Opensea support markdown in the description. With this I was able to create links to both the video and the images in the descriptions. It’s really nice.
Not much else to say here. I did some updates to the contract as well to support a “placeholder” URL before reveal.
Total time spent up to this point: 8 hours
My first real spend on the project was a $20USD domain. http://speedrunai.nft I bought this from Unstoppable Domains. What I didn’t realise is that the .nft TLD isn’t supported in all browsers. Only Opera…
So I bought another from Namecheap.
This is in use today.
Coming back to the images and word list, certain words like dust and erosion didn’t work to create interesting images. They are too abstract of plain. Also some prompts just plain sucked. Lol
In the end I discarded over 90% of the submitted words. Sorry guys.
I also spent a bit of time here chatting about the project. Including telling other mods to shush in the diary channel.
After another sleep I had a bunch of images and decided to use some for marketing purposes.
A logo and some sample images.
In hindsight, this should have been a larger focus of mine. The project was ultimately doomed because I didn’t do any marketing beyond this and shilling a bit on launch day.
Time spent so far: 9 hours
I’ve now spent as long in actual time as it took for the shark to generate.
At this point I’m pretty much done with items that need to be done manually. I didn’t want to wait for literally weeks for the image generation to run on my PC.
I rented a server from AWS that had a massive GPU and CUDA support. This reduced the generation time from hours to minutes.
Total spend on server: $37USD
I didn’t want to do any giveaways, but I did want to say thank you to everyone who submitted word prompts. Everyone who submitted was able to drop their ETH address and get a free NFT sent to them.
As the word collection was from a non-crypto server, this was a new experience for a lot of people, and a novelty more than actually valuable.
I set up the Discord. ~1 hour
Created the website. ~1 hour
Deployed the contract. ~1 hour. Twice (FML).
Switch on the sale for Series 1 and shilled in the shill channels of a few servers.
The collection was sold out in under 10 minutes.
Total time spent: ~20 hours
Total cost: ~0.45 ETH
Immediately after sell out there was a few sales on secondary, but then everything ground to a halt.
I tried my best to engage in the Discord but there was not a lot of engagement. People liked the art, but that’s all there was to it.
I offered 50% of all royalties and future collection sales to anyone who would come on as a marketing partner, but had no bites.
The project died without making back the initial investment.
I am sorry to everyone who bought in to the project expecting a profit. That was never what this was about to me and I didn’t communicate that well.
I have learnt so much since releasing this collection. These learnings will be carried into other projects that I work on. In fact, I already have.
You can still find the entire collection available on Opensea. Enjoy!
If you enjoyed the content please consider leaving a comment, sharing or hiring me.
Cheers,
Michael