Slaughterhouse
This piece is also based on the StyleGAN2 architecture, which offers a tool named “projector”. This allows the user to execute a search in the latent space - the set of possible images to be generated from the model. This search is made from any given image, belonging or not to the training dataset or even or the latent space. It will try to show the image most similar to the given one. For instance, if I use my passport photo as input, it will give me a face similar to me, from within its universe of possible generated images.
What happens, though, if the search image has nothing to do, visually, with the original training set? The system still tries to find something similar - but will certainly end up with something grotesque. This peculiar result may be seen as a blended image between the latent image space and the searched image. But in more exact terms, it is the image belonging to the latent space that looks most similar to the search one. If we think of GANs as a tool for visual artists, then this is one of the best ways to subvert its usage.
In the work created for the exhibition, I still worked with the Flickr Faces dataset model. In order to disrupt the tool, my technique was to feed the architecture with faces of animals. Since they bear features common to human faces (eye, mouth…), I supposed correctly that animal-like humans (or human looking animals) could be found. The search is a labor intensive task, since it can be run multiple times, with different random results for each input; the pieces are the result of my personal curatorship.
