How did you use ML and AI in Working For Each Other?

A: ML was used in the Te Kore video sequence (Runway), and in the synth tracks that accompanied some of the films (NotePerformer). I write them out like I would for real musicians, then NotePerformer generates a convincing output that articulates the music using ML if articulations aren't provided. Deep Dream Generator was used to bring stills to life, and for generative AI, such as the little skits about the eggs frying on the stove, or the No Disguises loop. Those are just short 5-10 second video loops that I put on the timeline. All the psychedelic colours are channel inverts. The basic loop has two parts: the initial clip, then the same thing in reverse. Been doing that since the AMPP satire skits. - Why did it work so well? I've been using DDG since 2018. It started with crude tools, and keeps improving. My view is that it's a powerful tool if you've got a strong visual sense and a handle on the kind of language that results in a successful prompt. It's not like you always get the output you want. Being over-literal helps. Sometimes I roll with it, other times I ask DDG to regenerate the prompt. With my comic strip, part of the fun was writing over the AI result. Paul's a surrealist, so we shared a love for all those crazy paintings and film loops. AI brings a different flavour that makes the film even more surreal. - The overwhelming majority of the film was still human-driven. I wrote and recorded all the songs and the tracks, other than Paul's overdubs. We cut it the old-fashioned way. There's probably a small amount of ML in Premiere Pro, but it still feels like it did when I was a teenager. I like to call it aleatoric cutting: I stack a bunch of loops over the base layer, then I refine them, much like the overdubs in a multitrack record. That way it's open to chance and happy accidents, which AI can still provide. Been working on that editing style since 2022. I prefer to see it as a VFX tool that works for me, rather than the end of the universe.