Available for SD1.5, SDXL and Pony! -> check version
This LoRA has been created upon request and I hope you all like it. It is the same principle as my other Ladder Cut-Out Shirt LoRA but another style. Check it out here.
It is based on a set of 23 photo samples and trained with the PicX real model. It works with front and side view as well as realistic and non-realistic models. With non-realistic models you might need a higher LoRA weight and a few more runs. Use a strength between 0.6 and 0.8 for best results.
Don't forget to share your images, I'd love to see what you have created. And if you have a few minutes left, make sure to have a look at my other LoRAs. Happy prompting!
Description
FAQ
Comments (6)
I will say it's a bit overtooned on both models I've tested. It really struggles with alternative styles outside of the base causing some artifacts in 1st pass. You also may want to try cleaning up your prompts / training for more traditional prompts.
You are adding tags like "blue, front, white, hand, one, woman's" which are corrupting the datasets without proper spacing.
instead of "anime, ladder cut-out shirt, beautiful petite woman, age 21, brown braided hair, bangs, Italian village background, brown eyes, irresistible, (casual:1.3), cute ideal face, ideal body, cute smile, shorts, (casual:1.3), zPDXL,"
I'd clean it up to " ladder cut-out, 1girl, brown hair, braided hair, bangs, brown eyes, smile, shorts" (for background, keep it simple like "godrays, wooden walls, night, clouds, etc"
Adding the filler doesn't help with the model.
@Bombalurina Wow, thanks for the detailed feedback. Did I understand you correctly that you are suggesting to reduce the amount of tags used when training the LoRA?
@BrokenTycoon More specifically, don't add filler words or useless tags. "age 21, age 4, age 100, etc" do nothing to change the image and muddy the training data.
@BrokenTycoon Also download the "Tag Complete" addon to get a better grasp of what tags work and don't work.
@Bombalurina Awesome, thanks for the hint! I will try to implement what you have suggested.