Automated T-Shirt Design

I am determined that this will be the year of completing projects. And to that end I am just finishing up on delivering a working version of my automated t-shirt design company. This projects holds a lot of lessons for me on just getting a project out the door and I’ll recount those lessons here.

Initially my idea was very simple. To look at social media and scrape concise phrases that can then be analysed and paired with an image as a t-shirt design. This design would then be sent through a drop shipping t-shirt company and finally marketed back to the original source.

Now this isn’t without it’s issues.

Taking an example, let’s say the system came across the phrase: “I love my Mom!” Natural Language Processing is then used to pull out the nouns and verbs and using what it finds it would perhaps find a picture of a mother or perhaps a heart. Stitching the text to the image it then ships the design off to a drop shipping company to create a product that can be sold and finally it markets that product back to the original source of the phrase.

Now this isn’t without it’s issues. Automatically creating t-shirts means that I can easily find a situation where I take something out of context, or perhaps the juxtaposition suggests something else. One example was the system pulled a joke from #oneliners: “Does this rag smell like chloroform to you?” Now that’s not the most sensitive joke there ever was in itself, but you can see where it’s going. Now in my tests my algorithm picked up on rag and in its searching and selecting came back with a little girl carrying a rag doll. It then paired the phrase to the image and voile I’ve offended everyone.

To combat this I’m not going to take the step and market shirts back to their original sources. I think I’m going to spend a few months just posting the t-shirts to my own twitter account to make a record of those ideas that are being picked up and see how offensive these are, or perhaps this was just a fluke.

Overall the whole project was a good learning project that did take a few years. Not so much because the coding was hard, but because the project was just larger than what could be accomplished quickly. This led to breaks being taken after a while, learning different systems along the way and also dreaming bigger than what I could accomplish in a certain amount of time.

The biggest lesson I learned was that if I reframe the question of what can get this project done by the end of the month, I can find pathways that are easier and refocus efforts to complete. While this is not a commercial project per se, if it was this would allow me to test the idea in the wild before committing more time to beefing up its underlying structure.

But the aim was to gain experience and lessons here and then move onto a project that looks specifically at crawling the internet for news (or using a service such as the GDELT project), writing news stories, arranging those stories against a back drop of supporting video clips and then publishing them on a platform like YouTube.

The aim is to automate the news, with a little luck I’ll get to start that project this year.