2023.07.01 update：added more pictures and real cosplayer pictures
==================================
Recommend weight 0.8~1.0
I followed the guide of rerorerorero and made this clothing lora. It's just an experimental lora.
I chose my favorite outfit from umamusume. The result is not bad, but it is easier to generate horse ear even without the horse prompt added.
Training info:
base model: anime full pruned
replaced costume tags with eishin flash cosplay
resolution : 768x768
rank=alpha=16
batch size = 2, epochs = 12
optimizer: lion, lr = 8e-5
Images set
96 face cencored images for 3 repeats
94 original images for 2 repeats
20 nude images for 1 repeats
4 focus images for 4 repeats
Description
FAQ
Comments (5)
what "ware" Block Weight?
ware:1,1,1,1,1,0,0.2,0,0.8,1,1,0.2,0,0,0,0,0 for clothing+face
Could we have the dataset if possible? I'm interested in it because this LORA comes with a nice style.
I added training data of full images, censored images and focus images. I didn't upload nude data specially because I generated it by stable diffusion.
@kazanesou434 Thank you so much! The dataset looks nice
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.