Update 5-25-2023

(Firstly, I couldn't decide between which one cuz they were both doing different things so I decided to give you both loras. One is very good at capturing the chibi style of the anime but it's a bit baked and saturated which might not work for some people. The other is more tame with the saturation but seems to love focusing on full body shots)

I completely replaced the previous lora with a new one. This one will allow you to get her in different outfits and stuff like that. Funny expressions works now too. Here are some sample prompts:

Her canon outfit: puffy short sleeves, white dress, striped ribbon, skull-shaped ornament, pantyhose, red and white shoes

(Funny Expression) - makes a funny face, generally ridiculous like they were in the anime. It's a great way to just get a random face expression that might be entertaining.

Blue lines on face - Attempts to add those blue lines you see on her face constantly when she's having a mood. Combine with blush for better odds of it working right. Can put it between ( ) to help get the prompt.

Overall it has a lot of room to mess with. CFG scales, additional network weight adjustments, prompts, sometimes the trigger word isn't needed, sometimes it is to get her body proportions right. Fullbody works now, it was finicky before.

I hope this will be fun for everyone! Post your art and let me know what you think.

__________________________________________

Use weight .75 if you want to mix her up with styles and other prompts, weight of 1 will probably work now per update with many outfits but not all. Up to Weight 1 if you want it to be more stylized but also makes it less flexible to work with styles and poses.

Higher steps seems to make better looking eyes and more canon to the anime style.

Play with weight .7-.9 to do various scenes and poses. You can write in "orange eyes" "red eyes" together if the lower prompt weight is messing with her eyes.

If you use the prompt "Funny Expression" and describe some basic face stuff like :o, :3 etc you might get a well, funny expression. I screencapped a ton of her facial expressions. I think the checkpoint makes a difference with what you can get with that however.

She is rather expressive with her face as a result of the images I trained with. Will need to mess with weighs to get her to do what you want.

Lastly, my training data contained a lot of hand shots, many very closeup. This lora has better odds of generating less cursed hands, but you're still going to get bad hands.