Is with a great pleasure that i present this Lora.
It's Composition Lora
ITS Panty pull Over-The-Knee
I recommend strength between 0.7 and 0.9
I recommend use the keywords:
otk_panty_pull, otk_panty_pull_with_one_hand
OR
otk_panty_pull, otk_panty_pull_with_both_handsDescription
FAQ
Comments (10)
on v1, both hands sometimes is a pain-in-the-ass to archive
Ehy! I was collecting images for this (forcing your previous LORA about a standing undressing), both F2M and M2F... Let me know if you need to improve the dataset.
which one? This or the stand undressing? That one is very overfitting.
spanking_cutie yes, not this one, I was talking about your previous, old LORA, about undress for spanking, but standing right... I started to force it to make the body bent forward, the succes rate was poor, but in the end I got several images for a dataset...
What do you mean with "overfitting"?
settima_ai overfit is exactly the struggle you sufering. Is when the lora hardly accept other things. even background style.
I was thinking in launch a v2 of that one less overfitting.
PM me, add me in discord i would like to see the images
spanking_cutie I've read your tutorial about how you create Loras, using self AI-made images as dataset. Is interesting because till now everibody told it was a bad way where AI amplificate all AI defects. But recent hi-quality Checkpoints like Illustrious probably changed this. Would you like to publish the dataset you used here? There must be a place in the "files" area to upload the zip. I see some creators publish the dataset, also.
settima_ai once you create the dataset on civitai, there's a download button, beside the auto-tag. Copy the link
(i think is that)
you can also publish the lora , and add the dataset to the files
spanking_cutie Sorry, I explained myself bad. My request is for you to put here your Dataset. I would like to see it, to understand how it's done!
settima_ai sure