This is my third try making a LoRA, so any feedback is still appreciated.
Im using the LoRA with a strength between 0.5 and 1.0.
Trigger Word: lowlegp
Works well with: groin, lowleg skirt, lowleg pants, lowleg leggings, lowleg shorts
Looking forward to your artwork.
Description
FAQ
Comments (6)
I can't tell you exactly what's wrong with the training settings or dataset, but this lora is way too destructive: https://files.catbox.moe/92paxs.png
It even breaks latent upscaling (causes image to become blocky at much higher denoising strength)
I would recommend going from default settings on the one-click installer for lora training, since I know those work just fine, and maybe trim your dataset to 100 or less images with a variety of content, artstyle, poses, etc, as equally distributed as you can, and 5~10 repeats.
Oh and train on NAI model, other models can give very broken lora. In fact if you're not using NAI that would be my first guess as to why the lora is broken.
I agree, it completely removes all the stylistic features of the checkpoint
Thanks for the feedback. Once I have enough quality training data I will make a V2. Until then I recommend the Add more Detail Lora which improves the quality a lot. I trained the Lora with the Kizuki Anime/Hentai checkpoint I can retry with the anything v4.5 maybe that helps.
In terms of ideas, many of your loras are undoubtedly a good fill in the gaps of NSFW, thank you and please keep it up!
Man, awesome loras. Can you please do unbottoned shorts or pants lora?
I second this. having trouble with getting consistent results for open/unbuttoned/unzipped pants, especially unzipped with top part folded to each side.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.