st0ne
Fri, 21 Oct, 2022
Algorithmic ritual
Experimenting with generating organic forms using a styleGAN3 model trained on an image-set of rocks (https://www.kaggle.com/datasets/salmaneunus/rock-classification). st0ne is a work in progress exploring ritual, blending organic material with digitally generative algorithms.
Choosing to use a GAN based model rather than the more commonly used diffusion models for the GAN’s textural qualities I wanted to recreate an archival and organic aesthetic. Projecting the visuals onto a lace mesh attempting to create and ethereal effect as the image parses through the mesh and onto the wall, layering the image on top of itself.
Accompanying the video is a piece co-composed with a mixture of raw audio generators (https://www.harmonai.org). Moving away from solely using Generative algorithms as producers of content, to a collaborative approach utilising algorithms to generate or ‘spawn’ (https://spawning.ai) new audio samples, based initially on my own field recording, that are then worked on by myself and re-input back into the algorithm in a cyclical back and forth between myself and the machine finally arriving at the final accompanying piece.
The input audio data to train the audio diffusion model (dance diffusion) is all sourced, with the creators consent, from a glitch/archival dataset, while the initial input audio samples are personal field recordings. As Ai or generative tools become more prevalent its important that the issue of consent of creators is acknowledged. For more on this I refer to Holly Herndon and Matt dryhurst who have been doing amazing work in this area, creating a platform to allow creators to opt in or out of their data being used in these training models (https://haveibeentrained.com).
Coining the term ‘spawning’ which is similar but distinct from ‘sampling’ they hope to create a dialogue that allows for these new tools to be utilised in a fair and ethical way where people can be reimbursed for the use of their data in these models. Creating a system that benefits all for their labour.
