BigSleep, GPT-2, Design Issues, JSX, and Adobe InDesign.

The images for this zine were created using the machine learning notebook BigSleep developed by Ryan Murduch in January 2021. While learning to work with this model I used a standard prompt “A Pair of Running Shoes by X” where X is the name of a noteworthy person in the history of graphic design. I generated a few hundred images initially. I batch renamed the images to make it easier to select images at random later on (this made using the title of the image in the captioning a little silly and uninformative.)

The text for this zine was based on a text prompt from me and then the rest of the text was generated using the OpenAI language model GTP-2. I trained the model on all of the available text from the journal Design Issues that was available on JSTOR. This method was messy. I downloaded the issue PDFs and copied and pasted the content. This does not correct for any data issues created by text set in multiple columns, captions, or running headers and footers.

I then used a JSX program I wrote to generate new iterations of the zine each time the program is run. The JSX script uses Levy Walker probabilities to generate most of the numbers used where InDesign has any room for variables. Text block locations and dimensions, type size, leading, catarted rotation, etc. are regenerated for each paragraph in the zine.