I have been working on training and producing the outputs of the gas mask training model for many years. It started in 2020 when I entered a South Korean AI art contest. The prompt was for something about “Faces”
My concept involved gas masks and clocks mixed together. These are two things with faces. I wanted to play with something that looked apocalyptic and had a concept of time. I gathered a few thousand images of found gas mask images and images of clocks.
But I also added some of my own images some of which I drew on dada.art some on paper with pen and then others with an oscilloscope by creating SVG files from hand drawn images.
Many thousands of curated images were used to train a SyleGAN 2 model on top of an existing public face model. The model takes a 2 to 8 hours to train itself.
When finished I looked through many thousands of images it generated to select interesting ones and used video generation software int Google Colab and Runway ML to create a latent walkthrough. This is where the software is used to move inside the matrix of the model through the placements of the pixels and colors and shapes to produce a morphing video.
These videos were enlarged and composted together for the AI art show. The art did not win the competition but in December 2021 it was selected for a special art exhibition in Paris with the Museum of Crypto Art called Crypto Art Revolution.
Here more about this on ArtBoxDMV soon.