Continuing the exploration of novel AI technology, I was curious to see what could be done in the realm of video. The concept was to create a zooming-in video that would ultimately conclude with the album cover graphic. The speed of the zoom would subtly change with the form of the song (verse, chorus, etc.), and at key moments, the visuals would relate to lyrical content in the song. Using Stable Diffusion Deforum, I was able to generate the new frames by auditioning the many available models until I found one that was good at generating the linocut style I was after. I was also easily enough able to guide the visuals towards the lyrical content. Unexpectedly, the most difficult thing was to make it all END with the album cover, instead of starting with it. Because of this, I had to create the entire video in reverse, and then reverse it back in Adobe Premiere. This made all the time cues very difficult to nail, requiring an extensive flowchart to map it all out. I made two iterations altogether, and although the second nailed the cues much better, the visuals generated in the first were my favourite, so I kept the first.