Lora came out a little bit overbaked, so lower weights are recommended. You can generate good images with high weight, but it will look more like sample image and less like your prompt. Overall it is okay, I managed to generate some nice images with it. It can generate pics from different angles, but main focus is on front perspective. Main problem, that bothers me a lot is that it generates poses from sample images. Haven't figured out how to make Lora remember only paw pattern on feet, but not the poses. Maybe I'll update it in the future..
This Lora can generate stockings/thighhighs or pantyhose. Main focus is on white legwear, it can generate black thighhighs, but it is hard to get good result for black color(not enough sample images).
Weight: ~0.5-0.7. Can be higher, but then it may ignore some of the prompt. Sometimes it can be good, for example, when you don't know what to generate, just put weight 1 and it will generate cool image with nice details.
Main trigger words: paw print, white stockings, white thighhighs, white pantyhose, black stockings, black thighhighs
Suggested prompts: from below, from behind, crossed legs, from above, on side, sitting, stepped on, leg up, close up, feet focus
Description
FAQ
Comments (1)
I had the same issue as well. What it turned out was the autocaptions on the sample images either had no pose captions, or was not accurate to what the model I trained on. I also had trouble naming the poses correctly so a good way to do this is to try to generate images on whatever model you're training on with the same pose. Once you have the right captions you can use those to label each image correctly. Hope this helps. Great lora nonetheless.