I’m someone who’s been creating generative art for a lot longer than LLM’s have been popular. Training and tuning your model used to be the vast majority of the work and it didn’t tend to yield great results. As a result, I’ve only used them twice before. I’ve written hill climbing algorithms, which are a form of machine learning, but it was completely different in both process and in outcome. My generative projects always use other techniques.

Personally, I find the vaporwave songs that just slow down a commercial jingle and throw a drum loop over it to be so much more creative than most of how people are currently using LLM’s and it’s not about effort. It’s about intentionality. The most intentional of prompts will have references to existing art styles and artists to fine tune what is desired. I think their most creative widespread use has been Midjourney’s Discord bot interface which allows you to iterate on a single prompt over and over to get closer to the intended result. Even then, the impressiveness is entirely in the technical details of how the art was made, not in the movement or in the art itself. It’s postmodern art with nothing to say about art.