HALLUSIONS
Artificial Intelligence
Generative Adversarial Networks (GAN) are a new revelation in Machine-Learning and not only understanding, but training Artificial Intelligence. The structure of these individual networks consists of dual neural networks that compare and contrast each other in a given training set. This technique learns the qualities of a given set of compiled images (e.g. landscapes or portraits) and learns to generate new data based on the statistics of the original. Through this process, a computer can essentially create an entirely original piece of artwork based on the information it’s given, as it adapts to realistic characteristics. ‘Latent Space’ refers to the void in between the two sets and this is often where you see the most warped, and abstract results.
Usually, the goal of running a GAN is to receive a large output from a smaller input, with the intention to receive realistic results.
RUNWAYML
I created a data set composed of 200 images of every graphic I created from 2014-2018. In order to create a video and generated imagery, I merged the group of images with a data set of landscapes to create a new set of textures. The result is completely abstract, and unlike anything I've ever made in my life, but in that way it's unique. In a way, this video demonstrates the abstract movement of my body of artwork.
Using my newly created data set, I merged it again but this time with 20 selfies of my face from different angles. I was hoping to get an abstract self portrait and see how the network understands my face in contrast with my artwork. I'm intrigued by the results, but I'd like to alter my settings more in order to achieve a more clear and definitive figure.
ARTBREEDER
I don’t have a strong background in Computer Science but I was instantly interested in this new way of creating artworks, especially because it’s still a new software. As of right now, I found two options available online; RunwayML and ArtBreeder. I wanted to experiment with combining two contrasting data sets to see how they would clash and what the result of that might look like. Some of the results came out as unrecognizable blurs whereas others generated contemporary, and original Ai-designed art which I hope to repurpose in the future.
