weight could be 0.6 - 0.8( makes it flexible for styles ) - trained for 768 ( may have better details) - SFW/NSFW
try any of these for default look - or put in Neg to reduce chance (making changes)
For appearance
Eris, blonde hair, blue eyes ,ponytail, black eyes,
is supposed to have gap teeth, or missing tooth, - however it may not prompt - can try (missing tooth:1.2), lips, - if there is a lora or style in future that adds gap teeth - pls do tell
for outfit
gold choker, silver tiara ,white skirt, cleavage, midriff, off shoulder, gold bands,
can add gold apple, - it sometime may spawn a K on it (supposed to)
Description
FAQ
Comments (9)
Nice model! I've been trying to make Velma Green (from the show) into a model for the past few days. The results aren't nearly as good as yours though!
well if you want i can give it a try, in my next set, (not the rest im going to upload today)
@True_Might Oh for sure. Yours would definitely be a lot better than mine. Here's a drive with the attempts I've made so far and the images and captions I used to make them, maybe it'll help.
https://drive.google.com/drive/folders/1wFywPu3h9do_iqEIBhRL3HwXAvvDYFze?usp=drive_link
Very nice.
The tooth gap is pretty essential to the character, I'd say. If you do a V2, maybe an extra dataset focused just on the head (ideally with a good amount of examples with the gap), that you merge with your main lora (I find a 0.7/0.3 ratio works best, main/head), will get it to show more consistently.
I had a similar issue with one of mine (the Mereoleona Lora), where the fang was essential, but it wasn't showing up consistently. The extra head lora was a really good fix. Making extra sure to tag it in any image it appears in.
so you your saying just face shots as a lora itself?
@True_Might Yup.
That's how I do characters.
One lora trained on full images, another trained on crops of just the face/head. Then merge the two at a 0.7/0.3 ratio. It gives some extra weight and attention to the head, and also makes the lora a bit more flexible overall. Don't worry too much about getting all the hair in them, it's the face that's important.
So you got one lora saying "yo, this is how the character overall looks when you use these tags", and another going "ay, and when you use these nifty face tags, this is how I understand them". Together they team up and do what they're best at.
ill suppose ill try it in my next set in the morning, theres only a few loras I have that could perhaps use this if it turns out well
didnt work out for me, all it did was make her teeth better, tried using any prompt that has anything to do with gap teeth. perhaps another time
@True_Might Yeah sometimes that shit happens. Back when I did the Tekuho Style Lora, I wanted to get the braces for different characters to gen well, and no matter what I tried it didn't catch on. I guess the images can only go so far, and the trainer has to have some kind of idea of what it's looking for based on the weights in the checkpoint (and whether or not it had examples trained into it), otherwise it might as well just be another activation tag.
Maybe I'll give the character a run in the future and see if I can knock a tooth out, if I can't manage it with training, I'm sure a knuckle sandwich might do the trick 🤣