Replace that hair with something different!
Place model in models/ultralytics/segm
It's pretty good but could be better. My fingers are hurting from all of the annotating. trained on 1000 anime images and 1000 realistic images. Does well with both image types but could be better. I'll continue to improve it. If you want to fine tune it further, I have a program I made on my Patreon that will allow you to annotate more images and train.
Also! I feel like there are not enough ADetailer models on here so if you need something specific made or if you need me to improve this one... Then let me know!
If anyone wants to be able to create their own yolo models with ease, I've created a program that makes it very easy to do! I linked the article for it in the resources area! It's very easy to use and I do charge $5 for it. I spent many long hours creating it and any support is much appreciated!
Description
FAQ
Comments (5)
Any tips on how to used this? I stick this into the comfyui facedetailer, and set it as the bbox and the seg set to this adetailer, but it seems to include the face anyway. Pepehmm
make a save image node off of cropped refined and send me the results!
I can't post an image here, but I uploaded a link here:
https://imgur.com/a/Vvwv1oG
Basically...it is including the face, or most of it, and the segmentation itself not that great even when it hits correctly. In your image guide it looks like it selects the hair so well, so...not sure what I could be doing wrong, but most likely a setting issue? Perhaps you have a comfy workflow?
@meowmeow12345 Got it. The problem is with the specific image or images that isn't working well with the yolo model. The Adetailer is including the face in every pass except one. That's a fault with the yolo model or the Adetailer Node (however you want to view it.) Bottom line is, I didn't train it well enough on images like that (possibly images with more hair than face?) If you want, you can continue to train the model on images like that using the easy yolo trainer I made! You could also make a node that outputs each inpaint sample pass. Krita is also an amazing tool that will allow you to inpaint the hair and change it using illustrious! Sorry I don't have a perfect solution for you...
Thanks a lot for this seg hair detector. 👍
For info is it possible to rename the file to give it an easier to recognize name? Because chances are that i ll forget what this detector do in a list of BBox and/or seg detectors to choose from with its current file name. 😂