@webmink I based it off the general idea that copyright of the input data applies to the output data no matter the license on the processing software: https://www.gnu.org/licenses/gpl-faq.html#GPLOutput

As far as I can tell, artificial "intelligence" image outputting systems do a form of sloppy translation, where multiple input images are translated into a single image that looks good until you look closer (although even the developers of such software don't know how it really work) - after all, such image generation programs clearly can't draw anything truly original (as far as I can tell, training an AI program to make it output original images would be like writing a random number generator that can generate endless random numbers with only non-random seeds).

It's well known that translations are a derivative work of the original work: https://copyright.uslegal.com/enumerated-categories-of-copyrightable-works/translation/