Fatum^2021–

Fatum (lat.): “In the eyes of the ancients, your fate was out of your hands, what happened was up to gods and demigods.”

Fatum is an escape from our everyday reality: from Covid-19, climate change, the rise of populism and nationalism, Trumpism, etc. etc. The aesthetics of Fatum are a study into how neural network models learn and progress. Can the audience feel the same thrill I feel when seeing a newly generated clip for the first time? The clips are fragments, glimpses into the trained model's micro-universe and its material essence, making the invisible visible.

The project started in May 2020, out of an interest in predicting the next frame of a video. I was inspired by things I'd seen on social media, especially Mario Klingemann's Twitter post of fireworks made with the pix2pix algorithm, and by a fascination with how it handled time and space.

With Fatum I work across different media: still images, short video clips, kinetic pop-up sculptures, 3D animations. Recently the focus has shifted to 3D modelling and animation, recycling textures produced by the modified pix2pix algorithm. Part of the work has also been to explore the glitches and imperfections of the algorithm itself, and to study how it predicts movement, first in kinetic sculptures, later in 3D animation.

In late 2020 I started collaborating with media artist Hannu Töyrylä, who programmed new features into the original pix2pix code and mentored me in image-making with GAN convolution layers. This made it possible to take microsteps through GAN epochs, mix convolution layers of trained models, and improve the next-frame prediction generally. I also began mixing different media in the same dataset, borrowing VJ techniques, going deeper into how a GAN image evolves.

Fatum predicts the future: it envisions the next frame. It relates to time in several ways, through the evolution and iteration of images, and by switching the roles of input and output. A playful exchange between epochs, frames and image sequences.