Endless Forms

Status: In Progress

Endless Forms is an investigation into combining machine learning output and traditional photographic processes. I was particularly interested in this idea of “bringing down the cloud”, or taking digital artifacts and creating physical objects from them. I used the botanical cyanotypes of Ann Atkins as the input for training a model, creating digital negatives from the output that I then reprinted using the same method.

I: Collecting the Input


Gathering the images proved more challenging than initially thought. I found collections of Ann Atkin’s cyanotypes at the websites of both Getty and the New York Public Library. However, due to the complex javascript the websites utilized and the need to human interaction (for example, in NYPL you had to manually scroll down to reveal more gallery photos) scraping these images were not a straightforward task. I ended up utilizing selenium alongside Beautiful Soup in order to mimic these interactions, which also allowed me to wait for any javascript to execute before scraping the DOM. I then saved the download links for each image in a csv file and wrote a third script in javascript to manually download each photo.

II: Training the Model

Now that I had all my images, it was time to train a model to learn from them and create new images. I decided to use RunwayML due to ease of use, using their styleGAN model. The output ranged from feathery delicate algae forms to painterly blocks. The writing too was interpreted and transposed as asemic writing on the output, an unintended side effect I find particularly interesting.



III: Bringing back down the cloud

title: “Endless Forms” —

I am now in the process of printing a curated selection of the output. To do this I first make a digital negative using photoshop and print it on Pictorico transparencies. Using the sun I contact print the negative on Revere Platinum paper applied with cyanotype emulsion from Bostick and Sullivan.