Assorted scraped dataset

DENIZENS OF THIS BARELY WORKING CLUDGE OF A WEBSITE.

I have one simple question.

Do you like monster girls?

Do you like anime baddies that would be liable to leave you desiccated, like the fake alien from a few years back?

Do you like it overbaked enough to have flavour?

If you answer yes to any of those, try my model, please, for the love of God. This slop website is burying my slop in other slop, and the measure of my worth as a human being and a GPU monkey is directly tied to how many downloads and posted images this thing has. We're all pressing buttons here, filming funerals and using the ones where the corpse starts dancing, so I don't know what to tell you.

This whole thing is assembled from a dataset I'd charitably call scatterbrained. The thing is a mess. I barely filtered for visually similar, a mix of a few obscure-ISH Pixiv artists who shall remain anonymous lest they find out and smite my Gaijin ass off the face of this earth. It isn't even hand captioned; I just autocaptioned it. I had buzz, didn't get training scripts on my Linux machine yet, and honestly, I'm just sitting here like

Like I'm sitting here with a GPU that represents a pinnacle of billions of hours of research and manpower and effort of mind and engineering, pushing pixies out the wall, and the crowning achievement of the technology is getting us to the end of history in the dumbest humanly possible way.

Filming funerals and watching the corpses occasionally dance. And all I can do is sit watching it, too smart to miss it, too stupid to do anything about it, and too comfortable on the couch.

Post-irony, post-humor, post-art, post-enthusiasm, post-reading comprehension, thinking, and philosophy.

And all I can do is sit watching it, too smart to miss it, too stupid to do anything about it, and too comfortable on the couch.

We're in levels of Dadaist absurdism that should not be possible. But here we are. Goodbye, punk. We deserve this crap.

Enjoy it, post generations.

A few short notes before you fire up your goonertrons.

Weight seems to behave well enough at around 1.0-0.8-ISH. 1.0 is honestly fine; just be prepared for it to have some idiosyncratic ideas about composition and backgrounds. If it is too active, just lower it slightly. I include this because long ago I stopped assuming you all even read these.

Trigger words, I've found, are largely optional. If your model gives you a bad time without it, try it. It seems to strengthen the output.

It honestly seems to be playing so far rather nicely with other LoRAs, but it might be my atrociously overcompensating ComfyUI workflow.

Have fun, and post your generations, please.