Concept/clothing model based on 60 images of simple stud barbell navel piercings.
The idea is to try to somewhat stabilize the hodgepodge that navel piercing generates by default in the base model, given that 9001 different types of piercing are all given the same tag. The result is... not as stable as I wanted it to be, but still pretty functional.
Usage
Basic
Works best at a strength between 0.8 and 1.0
Base trigger is navel stud, navel piercing
Issues
Has about a 10% failure rate wherein you will get the usual base model-type mess instead. Increasing hires fix denoise seems to help, but obviously comes with its own drawbacks.
As is usually the case with piercings (and you can see in some of my preview images), you will often get face piercings appearing when using ADetailer. You can workaround this by changing the ADetailer prompt and removing the piercing bit, but it can get pretty tedious.
Extras
In the interest of openness and community-helping, I have uploaded the dataset along with the model, this includes: images, captions, and the training json (for kohya_ss).
Description
trained on 60 images on NoobAI ε-pred 1.0
FAQ
Comments (8)
I also need navel piercings for most of my characters, and I quickly realized that the best way to do that is to add them manually with photoshop, it takes 10 seconds and the results are always better. no hate tho, your lora works and the concept is good but as an advice, from my point of view, for small piercings like these, PS is the way to go
If you want to put effort in, imo it's far easier to just draw a couple of silver/gold dots in the inpaint sketch tab 😅
But yes, this is intended as a one-click thing rather than fiddling around.
Is this an a1111 thing ? I only used comfy so far. Anyway, you got cool pictures on your profile, i like the one with the cat sketch ! Have a good day
@RemzDNB yeah it's in WebUI (i.e. A1111 and Forge), though I presume there's some way of doing it in Comfy as well.
And thanks, have fun generating ^^
From inspecting your tag, I think that the series trail tags was unintentionally split ei. raven (dc) -> raven, (dc). I believe the culprit is BooruDatasetTagManager, or did you do that intentionally?
I have noticed when tagging my more recent LoRAs too. I am pretty sure it's TagGUI that's doing it when generating tags. It shouldn't have any real impact on output given that it's generally just character tags that get scuffed, and those are not the primary purpose of the LoRA, but I do tend to go and manually fix them now.
If you use BooruDatasetTagManager, you can turn that setting off here -> https://github.com/starik222/BooruDatasetTagManager/issues/142 I have no idea why this setting was on by default.
@TinySandwich Aha, well that might help with other stuff too 😄
I have never had any need for tag weighting, yet they keep cropping up - I wondered where it was coming from. BDTM is one of the most bizzarely put-together programs I have ever used, but yet to find anything better 😅
Thanks 😊
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.