In last week's experiment, I produced some surprising and some confusing results. I went to the workshop by Runway's team and understood their process of making this product. I realized that to effectively predict the result of my machine generated artwork, whether it be on Runway or not, I need to know how to train my own models. With my limited knowledge on machine learning and current time constrain, I decided to hold my ideas for later.
A Coloring Book about January 6 Insurrection
I did this project with Francisco. It was inspired by the PhotoSketch model on Runway ML. He had an idea of making the January 6th Insurrection, a historic/ongoing event into an educational material, a children's coloring book. I thought it was humorous, even a bit ironic because our education system often interprets the history incorrectly, unfairly, or erases it all together. From my own experience as a semi-foreigner/new resident in the U.S., it's a daunting task trying to navigate and understand U.S. politics. Although I did watch the series of events unfolding live on television, it still felt fresh and exciting to hear a perspective from a journalist like Francisco.
In this book, we use actual editorial images, remove their texture, depth, and color content, leave only the outlines. I take it as a way to speak the truth, yet gives readers the power to render, imagine, and interpret this event.
We talked about how the storytelling should be. We could either follow a single character in the insurrection, or introduce the entire event documentary-style. We landed on the latter because we had more readily-available resources to do so. Francisco wrote down the bullet points in chronological order, and I began looking for editorial photos on Getty Image. I was able to find tons of photos with striking composition, very impressed by those photojournalists.
I began by testing an image in Runway ML, in Photoshop, in Photoshop AND Runway ML. They produced very different results. I used Find Edges in Photoshop and adjust the exposure to make it coloring book-like. However, even though the ML output was not ideal for coloring book, I did like its abstract and mysterious style. It almost decontextualized the image.
We complied 11 images and ended up making two versions. One was made in Photoshop with text. The other was done by PhotoSketch ML model without text.
The no-text, abstract version was given the name UNTITLE and all in Comic Sans. I wonder what kind of reader experience would this version gives out, especially for those who aren't familiar with U.S. politics, did not watch the news, or living in the future 100 years from now.