Carlos Tejeda Ocampo


Screams (2020). In 1982, when walking along the fjord, Munch saw the skies turned “blood red”. He described it as an “infinite scream passing through nature”. This would become the inspiration for “The Scream”, an impressive symbol of anguish, agony and anxiety. In these uncertain times, when the world and nature screams, we consider more than appropriate to revisit the iconic work of Munch in order to capture the current zeitgeist. Using generative adversarial networks and evolutionary computation, we create “Screams” a series of twelve reinterpretations of “The Scream” by Edvard Munch. Some screams are painful, others liberating. Some screams are abstract and etheral, others are felt in the flesh. Some screams are human, others animalistic. Some screams are abrupt, while others will haunt for years to come. Some screams are shout to the world, others are suffered in silence. But in the end, screams are, for the lack of a better word, important, since some feelings that can only be conveyed through a scream.

To create this art series, we employed a newly developed variant of the Deep Interactive Evolution algorithm (DeepIE). DeepIE combines Interactive Evolutionary Computation and Generative Adversarial Networks to allow for evolutionary controlled exploration of the latent space in a trained generator network. In this case, we employed the generator of a StyleGAN model trained on the WikiArt imageset. This way we end up with a limitless mine of potential artistic images from where to extract our artworks, as we have done in previous iterations of this workshop. However, this time we decided to take this process one step further with the development of StyleIE. StyleIE is a modification of the DeepIE algorithm designed to work in tandem with the Style-Based generator of StyleGAN. StyleIE applies the genetic operators into the disentangled styles vectors encoded in the secondary latent space of StyleGAN. This allows for unprecedented level of control over the evolutionary process on specific image features for generative models. DeepIE and StyleIE allow us to become both explorers and sculptors, carving our way into the intrincate depths of the latent space, searching fot the harrowing screams lurking on it. “Screams” constitude the first succesful art series generated using the novel StyleIE. The rest of the series can be seen at:
DeepIE was originally introduced in the 2018 paper “Deep Interactive Evolution” by Philip Bontrager, Wending Lin, Julian Togelius and Sebastian Risi. We express our sincere gratitude to these authors, who made possible both this art series and the development of StyleIE. StyleIE was developed by Carlos Tejeda expanding on the previous work done by the authors of “Deep Interactive Evolution”, for his master’s thesis. The abstract of said thesis can be consulted here: