Old Sights, New Visions: Controlled Uses of Diffusion Based Image-to-Image Translation for generative video

This study investigates controlled uses of diffusion-based image-to-image translation for generative video within the context of an arts practice. Image translation has proven to be a powerful tool in creating video art, but achieving both temporally coherent and visually diverse results remains a challenge with little formal research into the subject. Through a sequence of targeted studies of increasing complexity, this paper outlines the control techniques that were necessary to achieve quality results for images, animation, and video. I used my findings in the creation of two original art pieces, Kiss/Crash and Crash Me, Gently, which use these techniques to reflect on the nature of images in the age of AI.