I used cp decomposition to make the size smaller.
You may need to update your locon plugin to fully utilise this model!!!
This model may not be compatible with the AddNet plugin.
Apologies for my inactivity. I was experimenting with different hyperparameters.
Description
Trained on 99 images.
dim 1, conv_dim 1
FAQ
Comments (3)
It's a bit too rigid and doesn't work the best, unfortunately. I wish I could offer advice - there should be sufficient Miyu art to train a decent LoRA, but maybe not enough where she's in different outfits to keep it from being rigid.
I actually intentionally baked in the clothes. But I think the advice is quite valid: I used 91 official outfit vs 8 other as my training set. I will try ratios such as 40:8, and the default outfit with a trigger to see if it gets better.
Thanks for the advice.