HUGE UPDATE (Lora version coming eventually):
I got bored and created a bunch of images to train this with.
I have painstakingly and rigorously captioned every single image I've trained this with, and everything has been categorized into different concepts (While hopefully still retaining the old functionality)
So now this thing is a monster (at least in my opinion).
It's also definitely heavily biased towards my taste.
Main trigger words/concepts:
tensorfunk - Just generally combines characteristics of "Ye Olde AI Art"
ddfusion - Disco diffusion
mj_v3 - Midjourney (v3 and lower)
vqgan - VQGAN+CLiP
jxstch (My favorite) - JAX diffusion stitching.
Words that definitely have an effect:
microcosms - An attempt at nailing down that ultrafine detail you'd sometimes get with upscaling etc.
recursion - Anytime I saw something fractal-like, I'd generally put this word in the caption. It clearly has an effect on the output.
I am honestly pretty sure there are several more words. I'll look through the metadata later and edit this...
