This model is a lycoris for SDXL. It was trained on the sdxxxl_v30 checkpoint.
The model was trained with 39 pictures found online. The pictures were cropped to remove non-subject matter.
The captions for training were of the form,
"[extreme][close-up of] a women [lying|sitting|reclining|squating] with a gaping pussy,[small|medium|large] pussy gape,[small|medium|large][vaginal prolapse],[large labia|labia|anus|clitoris|clitoral hood|urethra]"
Any additional captions were added at the end and included details that stood out from the picture.
Examples: "shirt,high heels,on bed,breasts,jewelery"
Recommended Settings
The model might be slightly over trained and works well with a,
0.7-0.9 strength
2.5 CFG
25 steps
DPM++ 3M SDE Karras
Other setting may work as well these gave good results.
Strengths
Really good at extreme close-ups and close-ups.
The model can change the position and the locations.
Can generate very realistic anatomy.
Prompting vaginal prolapse works well.
Weaknesses
The model struggles with nipples. It might be confusing nipples and clits.
Has issues generating teeth.
It is not great at faces, prompting different a hair color helps.
Version 3
A new version will be released if the model can be improved.
The model would benefit from more full body and face images in the training data.
Description
FAQ
Comments (2)
Even though it doesn't necessarily include a gape, it definitely improves the details at lower body. However it messes up nipples and face, even on a newest version. Which is very unfortunate because I love the positive features it provides. Is there any way to make it to not affect face and bust, and only work in lower half? I'm not 100% knowledgeable on how loras/embeds exactly work and if you can tweak it to work around specific areas of an image.
Downloaded this a while ago and only got around to testing it now. This is a great lora and one of the best I have used for Asian women. I just dial down the strength for the initial picture and reapply it at a higher str using Adetailer, works perfectly. Thanks!