The tools and techniques described here were pioneered in 2021 by a diverse and distributed collection of amazingly talented ML practitioners, researchers, and artists. The short version of this history is that Katherine Crowson (@RiversHaveWings) published a notebook inspired by work done by @advadnoun. Katherine's notebook spawned a litany of variants, each with their own twist on the technique or adding a feature to someone else's work. Henry Rachootin (@sportsracer48) collected several of the most interesting notebooks and stuck the important bits together with bublegum and scotch tape. Thus was born PyTTI, and there was much rejoicing in sportsracer48's patreon, where it was shared in closed beta for several months so sportsracer48 wouldn't get buried under tech support requests (or so he hoped).
PyTTI rapidly gained a reputation as one of the most powerful tools available for generating CLIP-guided images. In late November, @sportsracer48 released the last version in his closed beta: the "pytti 5 beta" notebook. David Marx (@DigThatData) offered to help tidy up the mess a few weeks later, and sportsracer48 encouraged him to run wild with it. Henry didn't realize he'd been speaking with someone who had recently quit their job and had a lot of time on their hands, and David's contributions snowballed into PYTTI-Tools!