Put the model under models/adetailer
Only tested with realistic and semi-realistic images. There is no anime / cartoon images in the training data.
Description
All 4.x versions are mainly focused on fantasy themes. If you're just want to inpaint regular humans, the stable version might give you a more accurate result.
FAQ
Comments (19)
This helped fix some lost detailing during image corrections, and the current image is much better
dose this work for comfy?
If it supports Yolov8 format, it'll work. I've heard from other people that it works, but since I'm not using Comfy, I can't confirm.
of cause it works, why not? save to models\ultralytics\bbox
Thank you for the presentation.
For sideboob, it often produces excessive nipples. If you can fix that, it would be great.
What do you mean it produces excessive nipples? It's detection model. It doesn't produce anything.
@Jellon Oh, I see. Next time, I will add "sideboob" to adetailer, and see whether the problem is solved.
If you have more tips(like other prompts) please share with me.
@daixiaoluck261 The inpainting result is entirely based on the SD model you're using. And in that regard it works like normal inpainting. So if you're struggling, inpaint the image normally and then use the prompt and settings for ADetailer.
All my model does is detect the area which should be inpainted.
Is it suitable for anime?
I don't know, you can try it but as said in the description there are no anime images in the training data.
yes, i'm try v4.2 on anime, it can fix breasts and nipples.
@Jellon Tried it. It didn't work，checkpoint is waiANINSFWPONYXL v50
@feng_1 Yeah sorry, then you have to look for a different model that is tailored to anime. This one is really all about realistic and semi-realistic images.
what is this used for?
It's a detection model for the ADetailer extension.
does this work for XL or pony-XL?
works on pony realistic models for sure. Not always perfect, but mostly awesome!
Detailers don't really care about the rest of your workflow / base models. The ultralytics component is just a detection layer - it's trained to look for breasts here, or faces, or eyes, or whatever. Depending on the detailer, it'll either generate a bounding box (read: the thing you're looking for is in this rectangular region of the image) or a complex mask (read: the thing you're looking for are these pixels specifically, not just in a generic box) - and then it'll hand that mask off to whatever model you want to use. SD1.5, XL, Pony, Flux, etc - it doesn't matter at that point.
@kurokuro thx!