I was honoured to participate in the Performative AI: Neural models + agents workshop. This fabulous course was taught by Carlos Navarro and Harry Moller as part of Digital FUTURES 2022/One Planet.
I started with my archive of analogue photographs; landscapes and seascapes taken across many countries. I arranged 1000 of the most interesting into a dataset. These were used to train a StyleGAN model over 3000 steps. It was fascinating watching the initial faces erode into landscapes.
I used some of these images as prompts for the Disco Diffusion AI, alongside a text prompt of “A labyrinth of paths through an unquiet landscape, next to a tumultuous sea, John Virtue, Derek Jarman film stills, Unreal engine.” The images which worked best for these experiments had medium to high contrast to generate the required depth in the 3D models in Houdini and Blender.
The landscape in Unity made a complicated terrain, difficult for the agents to navigate. This resonates with the labour of walking on the remote paths where the initial images were taken. This has been a fascinating process, with huge thanks to the instructors for their fabulous work.