This is a generator of a running joke in the ru Touhou community. There was a challenge to edit a pixel art of Cirno by @motijan to make it resemble other characters. So this model is trained on that dataset
Images was tagged by DeepBooru interrogator with minor edits (maybe i'll tag them myself later on)
I suggest generating images with lora weights from 0.6 to 0.9 with CFG 12, to make it less aggressive bcs it WILL generate strange results. Also it seems my LoRAs are creating artifacts like blur/chromatic aberration/etc when using VAE. pic related
Shoutout to @galina_tsubakurovna
P.S. if low step models will be proven good then maybe i'll release them too