Nipple LORA for ADetailer / ADetailer用の乳首LORA
for Pony Diffusion V6 XL
required: nipples_yolov8s.pt
https://civarchive.com/models/490259
NP21
How to use
put <lora:PDXL_NP21i:0.6> in t2i and ADetailer prompts (large nipple size)
put <lora:PDXL_NP21i:0.6> only in t2i prompt
put <lora:PDXL_NP21i:0.6> only in ADetailer prompt
Strength: 0.2-0.6
使用方法
t2iとADetailerのプロンプトに<lora:PDXL_NP21i:0.6>を記述(乳首大きめ)
t2iのプロンプトにのみ<lora:PDXL_NP21i:0.6>を記述
ADetailerのプロンプトにのみ<lora:PDXL_NP21i:0.6>を記述
Strength: 0.2-0.6
NP12
"<lora:XLP_NP12:0.4> nipples" in the ADetailer prompt field
<lora:XLP_NP12:0.4> nipplesStrength: 0.4-0.6
Detection: 0.6-0.7 (coordination required)
日本語
必須:nipples_yolov8s.pt
https://civarchive.com/models/490259
ADetailerのプロンプト欄に「<lora:XLP_NP12:0.4> nipples,」
・貧乳なら0.3か0.4
・巨乳なら0.4-0.6(0.8までいくと左右の乳首の大きさが不揃いになりやすい)
Detectionは0.6-0.7くらいで(要調整)
Description
FAQ
Comments (7)
How effective is this LoRa in real style?
cool idea, maybe we need more ADetailer models for the small details :D
I think it is the best value LORA, but I feel that the size of the left and right sides tend to be uneven.
Does anyone have any good ideas? For example, would the problem be solved if the detection model were to detect the area including the left and right nipples as a whole?
In the prompt field for ADetailer, enter " <lora:XLP_NP12:0.4> nipples, "
・If you have small breasts, 0.3 or 0.4
・For large breasts, 0.4 - 0.6 (if you go up to 0.8, the nipples tend to be uneven in size)
Detection should be around 0.6-0.7 (needs adjustment)
You might be denoising too much. Try a low value like 0.2-0.3 and see if it's enough.
Can you make one for hands? I've been having trouble with hands adetailer.
bit late but if that's the case i'd try another model for your hand corrections. pony is notorious for bad hands. but the good news is that you can get them in 1 or 2 gens using a good SDXL fine tune, Animagine or Illustrious-XL. give it a try. go to the ADetailer dropdown > Inpainting and check the "Use seperate chekpoint" box, select the model you'd like to use and you're good to go. the upside is the other models are much, much better with hands (less "fiddly") and can them better in less tries than pony, with the tradeoff of 35-40 seconds of loading models into VRAM. I suggest the above because given pony's namesake focus, a significant portion of the finetune dataset featured subjects that simply dont have hands. same reason you get lotsa jacked-up lower legs/standing on tiptoes/why the tibia to femur ratio is whack. people who care about accurate hands are just going to use a different model. horses for courses, if you'll pardon the pun
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.






