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
FAQ
Comments (18)
will you make an SDXL version?
It's not dependent on the SD version. If ADetailer works with SDXL, this model should work too.
ah ok, thanks for the clarification! I'll try it out!
it seems adetailer is for automatic1111 only. I'm using ComfyUI. any plans to make it for a ComfyUI equivalent? there is the comfy-ui-impact-pack custom nodes that feature a detailer, for example.
@sandro609 Sorry I don't know about Comfy. It's only for ADetailer.
@sandro609 @Jellon This can be used in SDXL. You can load it in the UltralyticsDetectorProvider node and connect it to FaceDetailer or another detailer node
@aadsgdgadadggda ah right! thank you for the clarification! cheers!
honestly, this is a brilliant idea if we could put in the type of breasts in the adetailer prompt as well to guide it. Or is that the idea?
It would be possible, but currently ADetailer doesn't support that to my knowledge. Plus it doesn't really matter as you can just make a model for each type as you'd need training data for all types anyway.
txt2mask?
Small breasts are not really recognized. But thank you for your work đ
Can you provide an example? It does recognize small breasts in my test.
Here's an image I took from https://civitai.com/models/25727/jk-small-breasts-lite
I'm having problems getting it to detect; it often doesn't detect the breasts, or only detects one out of a pair. I'm trying turning up the detection confidence threshold.
You need to lower the confidence threshold in ADetailer. If that doesn't work, I'd need an example.
@Jellon Ah, I thought you had to increase the confidence threshold to get it to detect better. For reference, what do you set that threshold to?
@LarryLePoop No the confidence threshold means if the confidence is over the threshold it detects it. So the higher the less likely the detection. Generally I'd put it anywhere from 0.3 - 0.5.