Tries to replicate the look of (some of) GrAdm's works.
Please support the original fellow slopper on, uh, Civitai? Rule34? Idunno.
Description
Initial release.
FAQ
Comments (4)
Looks cool.
Also, up for art style suggestions?
I'm both lazy as hell and rather picky with the styles i like.
That said, however, i'm very much open to suggestions.
Very solid work!
However, as a "progenitor" for this lora's style, i did some testing and concluded that it works best at around 0.6 strength
Putting it at 1.0 causes some weird additions to the outfits, some of which persist even when using natural language to negate them (for example: Kotohime constantly came out with the flower patterned clothing and her junihitoe, even when i asked for a different outfit. After i specified the outfit with natural language and put weight to said outfit, this side effect was mostly negated, but sometimes said artifacts persisted. Only when i lowered the strength to 0.6 did it fully stopped being an issue)
Perhaps it is due to large amount of pictures the lora was trained on, but i'm not sure (iirc 98cuck told me that recommended amount of pics is somewhere around 80 pics, but he wanted to try to train a lora on 100 pics and above so i've provided 113 pictures for it)
i've used WAI-ANIMA v1.0 for said tests
Odd, can't think of anything that would cause that on the LoRA's end, more so since Kotohime isn't tagged as such in the dataset. I have also determined that the number of training images doesn't really affect the quality of the output (unlike with Illustrious), both from my own tests and those from a friend of mine, so i doubt that could be the cause either.
Oh well, whatever. As long as it works.