Another slightly difficult one that I'm very happy with. I still have a long way to go with LoRA training but I hope you enjoy this model! Used kohya_ss gui.
I am most likely going to retrain in the future to put more emphasis on the snaggletooth/fang and her gun.
Here are some parameters and helpful keywords:
• Checkpoint: AbyssOrangeMix3
• Trigger Word: galil ar \(upotte!!\)
• Weight: 0.6-0.85 but 1 is fine.
blue hair, brown eyes, animal ears, dog tail, serafuku
( i'll add serafuku, animal ears, blue hair and dog tail to improve accuracy, can also add green skirt, white shirt )
( Try adding detached tail to negative prompt if tail is incorrect, move serafuku to negative prompt to help with other outfits )
She likes to fiddle with things, adding objects for her to hold can produce interesting results. Some training images included her holding bottles with decently drawn hands. Guns are difficult right now, maybe try "holding gun" alongside a model that improves guns or check the sample images for tips.
Try other outfits too: swimsuit, bikini, kimono, maid outfit, sportswear
To produce similar results I recommend EasyNegative, this negative embedding was used in the sample images. I'm also using Clip Skip 2 and hires fix.