As this is pretty much an alpha any tips or feedback would be greatly appreciated.
A Lora trained on images of paizuri with multiple angles.
wights 0.8 - 1.3
recommended tags: paizuri, multiple_views, penis,
optional: multiple_girls, multiple_boys, (or specify like 4girls i think it might help but i haven't tested it much. the same applies for: multiple variations, different times).
I'm not sure if specifying the angles matters.
using a paizuri Lora may help.
currently can give okay-ish results, it seems to be able to recognize it's the same character and even sometimes makes differences. Can do something similar to "instant-loss" where it shows the girl (in this case getting ready for paizuri) in one panel and then the other is said action.
not sure what the best weights or width/height are. so far 512x768 or 768x512 seem to be working but need to do more testing.
Description
FAQ
Comments (7)
What does the :>= in the prompt mean?
i think it's a face, it was a tag i got from an auto tagger. usually those symbols are faces though.
The emoticon that shows the lips being tugged forward on the backwards pull whilst sucking on something, as seen from an overhead or semi-overhead viewpoint. Typically this means fellatio, but it can also be seen when a character is sucking on a more innocuous object, such as a popsicle or pen. (got it from danbooru wiki)
thank you!
When I used it, I thought this LORA was meant to adore WAIFU.
Enjoy WAIFU with various expressions in landscape, and have them PAIZURI in portrait. It's such a dreamy LORA.
Did you use NAI as the base model? I'm always curious what model is best for training concepts, lora looks great btw!
yeah, most Loras seem to be based on NAI recently AnyLora has been getting some as well