Trained on the style of the author of Prison school. https://danbooru.donmai.us/posts?tags=hiramoto_akira
Data is mostly illustrations with very few screenshots or manga panels.
Description
FAQ
Comments (10)
💕👍🏻
i'm getting mixed results depending on the checkpoint I use here, sometimes it looks like the artist sometimes not :(
It was not tested on any other models aside base anima. Merges/further finetunes will eventually not work depending on how they're done.
They might be overfit to death, preventing further alteration. Or requiring an equally overfit/overtrained lora to work.
@Kinky_Caterpillar tested on anima base too and barely looks like the artist style either, not sure whats missing
@Petamequier123 Don't forget that any quality tags/negative can influences the outputs.
My gallery example have metadata and are not cherry picked, you can try to reproduce them
@Kinky_Caterpillar Well I tried the remix option here with one of your images too. Not sure how other styles are more flexible or easier to use 😅
could you make the anime style too?
it has problems with expressions, it makes characters look angry regardless of what you prompt even at higher strengths and even when paired with multiple other loras, even with such that tweak expressions. Data must been insufficiently tagged.
You mean the eyes that are "tsurime" ? That's the author artstyle, rarely drawn other type of eyes. But expression itself as "smile", "happy" "smug" still work properly. And the eyes still can be influenced with the right tags such as "tareme"
@Kinky_Caterpillar oh!, oh! buhahaha, i was using tag tsurime for eyes out of own volition not knowing that it was reinforced in dataset, well that would explain the bias.