Hand Segmentation for ADetailer
Credits to mnemic for this article and Anzhc for this ADetailer model (see for more information)
Installation:
Download the zip archive
Extract the model into your ADetailer model folder
For the WebUIs like Auto1111, Forge and SD.next it should be in models/adetailer
For ComfyUI it should be in models/ultralytics/segm
Start it up (full restart if already had it open) and select the model from the ADetailer tab/node (tip for webui: you can go to Settings→Adetailer and change the number of models that can be used up to 10, very useful if you want to use more than 2 ADetailer models)
Make sure to have correct Detection thresholds/inpaint strength and so on for your images
Generate them hands!
It detects hands good enough for what SD hands can be. I might work more on this in the future. Please tell me what you think. Thanks! 😊
Description
YOLO v9c test version with v2 dataset
I have no idea if it's better or worse. You'll have to test it. Theory says YOLO v9 > YOLO v8
FAQ
Comments (20)
Does V2-V9C have bigger training data set or just the same set with YOLO9 instead of YOLO8? Great detection tho.
RemTheBestWaifu V2 means my dataset version, not necessarily bigger, more like modified for better training result (e.g., better masking). YOLOv9 or YOLOv8 is the network that it is based on. The v9c network architecture is in theory better, (but it needs more compute from what I found with the tests I did, I don't like it TBH, I'd prefer YOLOv10 or later, but those are not supported yet maybe? I don't know, my pc is cooked atm, so I can't really check effectively)
BBox version please?
KoujiAI BBox version comes with the adetailer extension/node (or should at least). Link for them here
Pitpe11 I already have adetailer, im not sure what you mean by there being a 'bbox' version just by having adetailer? Its either segm or bbox, unless its a custom fork which allows for segm model like yours to be treated as bbox?
KoujiAI As I said, the adetailer extension for SD WebUI by Bingsu comes with a bunch of models (list). Bingsu made the extension and trained the models when it was first released years ago. The hand_yolov8n is the BBox version of my hand Segmentation model. Mine is based on that model, it just does segmentation instead. Idk if they come preinstalled for comfyui so u can get them from here, and put them in their respective folders. Models are either Seg OR BBox, the result depends on training. Any model that has "-seg" should be a segmentation model. No custom forks or anything.
could you help me how to use this? i can use your eye detailer but im not sure how to use this as there's no hand detailer just like the face detailer in comfyui. pls help me, and tell me just the simple things i needed in order to use this in comfyui
hanada_civitai Make sure the model is in models/ultralytics/segm
and as I've told you in another comment:
You need these:
https://github.com/ltdrdata/ComfyUI-Impact-Pack and https://github.com/ltdrdata/ComfyUI-Impact-Subpack
Make sure to read the instructions!
Nodes: UltralyticsNetworkProvider → SEGM Detector (SEGS) → Detailer (SEGS)
These probably still apply, I haven't been able to use comfyui since my pc is non-functional right now, so if this doesn't work then I guess you'll have to figure it out on your own. I'll update if needed when I get my pc working again. I might make a general usage tutorial for these in the future.
Amazing !!!
Hello, very nice work! I use it often and it works quite well, almost without false positives, much better than the basic Yolo models. However, it is quite useless for mechanical hands or cyberprosthetics. Could you add some mech images for the next version? If it is compatible of course.
@OtakuFra I'll try to add some mech hands sure.
bug report: false detection of feet will occur
I don't know what i'm doing wrong, because it correctly detects hands, but changes absolutely nothing in the output
detects feet as well
Increase the detection threshold.
Amazing model! I noticed that some people complain about foot detection, and I wanted to ask whether it would be possible for you to make a separate model specifically for that. Sdxl tends to break legs just as often as arms, so it seems like this could be quite relevant xd
p.s. I can detect feets using this model with a low threshold, but I’d like to be more confident in the results...
@degurshaft Would if i had time
Flawless! And thank you for clarifying that it's segm!