This is a Furry/Yiff LORA trained to emulate the style of the artist "Jishinu" with about 4K Steps.
It's really good at the style itself and can even replicate certain characters pretty well, thanks to the dataset.
Multi character scenes also generate nicely.
I recommend using at 0.5 strength.
I hope you will enjoy this model!
Description
Version 1.0
FAQ
Comments (2)
Would it be alright if I asked what kind of hyperparameters you used for training?
e.g. learning rate, lr scheduler, weight precision, bucketing, xformers, max res parameter, and how many images it took to get this quality?
and if you used a certain subset of years in the pruning of your data?
and what the base model was that you trained on?
- fluffyrock
- fluffusion
The model is amazing!
First of all, thank you so much for the compliment! ^^
For my LORAs i like to use constant scheduler, learning rate at 1.0, of course with xformers as well.
I will say, to get something that mimicks the style well, you'll want, at the very least, a dataset of about 100 more or less consistent images, preferably with solo character focus.
But the more the better (typically ^^)
The right tags can also make a world of difference and i did have a couple of upload dates specifically tagged for better reference, both for me and while learning x3
It was trained on the pretty neat Furryart-FRv2 Model, which itself is a merger of multiple models i believe, and helps greatly with artist style impressions.
I hope this helps you a little and i recommend experimenting every now and again, until you end up with decent results :3