This is a style model that gives softer faces to characters, this was a mix of several artist that although not doing what I originally intended turned out very cute. Tends to work best at 0.6. This model was trained on NAI anime-full-pruned and generated in AOM3. Made with the Kohya tool for AI Training through google colab. Any suggestions to improve or recommended things for the data set would be appreciated.
Description
Details will be added when tested further, this version is more concentrated with flipped images like version 1 but does have several artist applied at different weights making it not as diverse as version 2. This was created mainly to be used with Soft n Fluffy Parka Lora.
FAQ
Comments (6)
for some reason, breast size is messed up when I use this. In future releases, could you remove breast tags from the image tags during training?
This is my first attempt making a style lora and also my first time dealing with so many pictures so it makes sense that size would be affected due to the dataset. Can you specify the specific size that's messed up is it the generic 'breasts' tag or is it a specific size like flat/small/medium/large/huge/gigantic. If its a specific size I'd need to add more of that size, if its the generic then yes I'd need to remove it from all pics since all of them have the size labeled anyway.
@rulles its generic sadly. if you want to, I can show how to use WD14tagger which can easily fix your dataset
@silverace71 there isn't a trigger word for this Lora but if it works without it you can message me from the discord server (same name and profile). Otherwise I'll be sure to just remove the generic 'breasts' tag for future versions version since all the pictures don't need it.
@rulles whats the discord server?
@silverace71 Just the civitai discord. This should work: https://discord.gg/UwX5wKwm6c