Hi! I'm experimenting with training parameters to make my style models as small in weight as possible so here goes 55 megabytes Ayanakitori style lora. I think it performs very well.

Every training image had bar censorship so in order to avoid it you better put "uncensored" to positives and every single censoring tag (censored, mosaic censoring, bar censor) to negatives.

Any advice and feedback are extremely welcome as always!

I will be very happy if you make a post and share your results!

Thanks for visiting and have a great day!