Low size, low pollution, flexible Princess Peach LoRa trained without clothing bias (especially no crown).
If you like, please post your pictures and leave a review!
Follow me for similar loras in the future!
Suggested Use
Base Model: Any!
Weight: 0.8
Trigger: None!
Extra Tags: lips, ponytail
Known Issues
Eyebrows sometimes are imperfect.
Her mouth is always open.
Hair sometimes imperfect, some pollution.
Weight not rebased to 1.
It is hard to find suitable images for training. When I find more and make v2, these may improve.
Description
FAQ
Comments (5)
So I understand it's without bias for it, but does it at least know how to do her usual dress and crown? I'd honestly rather the option than a lockdown to either.
No dresses in the training and only one image had a crown. So only if your base model knows it. I tried to train her different dresses before, it's underwhelming. For high accuracy you lose flexibility and freedom. You only can make good pictures of very common things, and sometimes a bad style bias because of low training data.
Even if you want that it's hard to get high accuracy on everything. You can have good dresses, but crowns will be bad because every training image has a different crown. Things like that. If there's consistency (like peach wedding dress model on my profile) it can be good, but it's rare in training data. I think there's already a peach model here that's trained on consistent official images, but it has strong style and other bias. I also tried to make multi clothes models, but it always loses accuracy from clothing that needs it.
Sorry for long answer for a simple question.
Amazing model! Thanks for sharing.
For this point: "Weight not rebased to 1" - How do you do this?
Small file size - How did you get it down to such a low size?
I don't know immediately how to rebase weight, I see that other people do it so I only know it's possible.
The small file size isn't hard, just set it low. I can do it that low with this because it doesn't have to learn much. Just hair shape, color, earrings, very general eye shape. The rest is base models or styles to fill it in, and that makes it flexible too.
A so adorable princess! WIth 2mb file. You're a hero!


