Makes card queen like girls.
Seems like it may be a little undertrained. Might need face fix to look good. Works better using 2girls, on stomach for lying on floor poses, and standing, upside-down for standing poses. all fours works too, for a both girls standing on floor pose.
Trained on 32 images, on 512 steps.
Description
This is a test to see if reducing dim to 8 helps reducing the artstyle influence. If it does, that means i can be less picky about artstyles, which means less overtraining.
It seems to have worked.
Like before, it needs a 1.5 strength.
FAQ
Comments (5)
For lying on stomach pose, it kept generating a 3rd person in between. Any tips on how to avoid that?
How much lora strength are you using? I often have to use 1.5 on this lora.
@fominhavi 1.3, but I switched to 1.5 and it still does that. Gonna have to experiment some more later.
@vqt Maybe it's the prompt. Are you using multiple girls, or anything else that would change the number of girls other than 2girls? It's the only thing i can think of.
@fominhavi I don't think there was anything in the prompts like multiple girls that would make it generate more. I've even put 3girls and similar tags in the negative prompt.