Hi, this is LoRA of Naaza from danmachi.
also this LoRA is my first LoRA so its not perfect but doing quite well job.
v1 trained with 95 images of Naaza and i used hassaku hentai model v1.3.
for v1 use weight around 0.7++
v2 trained with 44 images of Naaza and i used sd 1.5 base model.
for v2 use weight around 0.8-1.0
v2 better at drawing default clothes and tail. And more flexable.
Trigger words;
naaza erisuis, hair between eyes, dog ears, dog tail, dog girl, light brown hair, purple eyes,
for default clothes you can add naazadef, scarf, multicolored clothes, multicolored shirt, long skirt, black skirt
v2 has higher chance to generate outfit.
for red dress you can add naazadress, strapless, red dress, long dress, strapless dress,
Description
initial version
FAQ
Comments (3)
95 images, yet just 13 steps... something is sus here...
also go Dim 32 alpha 16... no loss in quality yet the size of lora is half of what you have now
and preferably all anime characters should be trained on NAI model anime-full-pruned
13 steps per image and 3 epochs. Also thanks for advice its my first lora
@beyyaz6384 the steps in that box count as total amount it took to make the lora, not per image i think, at least thats how ive been using it this whole time