Put the .pt model file under models/adetailer
Commission for @AikaKittie - Thanks!
This is trained on YOLOv8-seg - the instance segmentation model, an aDetailer model trained to help detail the eyes, eyebrows, and nose area, basically the area a mask would go on the face.
Trained on 322 images - overall pretty consistent - trained mostly on humans and minimal training wearing actual masks, - sorry furries - this is technically the 8th iteration.
The example images are merely showcasing the ability to detect the mask area, it is technically prior to the aDetailer being used on it. When using it on a tall image I suggest using a separate width/height to either be wider or square.
Description
Put the .pt model file under models/adetailer
Commission for @AikaKittie - Thanks!
This is trained on YOLOv8-seg - the instance segmentation model, an aDetailer model trained to help detail the eyes, eyebrows, and nose area, basically the area a mask would go on the face.
Trained on 322 images - overall pretty consistent - trained mostly on humans and minimal training wearing actual masks, - sorry furries - this is technically the 8th iteration.
The example images are merely showcasing the ability to detect the mask area, it is technically prior to the aDetailer being used on it. When using it on a tall image I suggest using a separate width/height to either be wider or square.
FAQ
Comments (19)
I am SOOOOoooo excited for this!
Heya - I'm using RunComfy and I don't have the usual 'Ultralytics' folder to upload this into - any idea how I can make it work, please?
I also don't have the Adetailer folder
@andrew_j_ I don't use Comfyui but If it's anything like adetailer and the instructions are asking for a folder that isn't there, could try creating the folder and putting the file in there and restarting
@andrew_j_ The way you get the directory in comfy is you download a yolov8 model directly from your manager and it will make the directories you need
I use impact pack and impact pack subpack for adetailer, download them from comfyui manager should generate the folders. Add adetailer to ultralytics folder and use the ultralyticsdetectorprovider from the subpack to load into memory.
imagine if this could somehow be chained with reactor so that it wouldn't overpaint the mouth.
there is a mask occlusion node that now works with reactor, more fiddly than this but it works.
n67113610 node as in comfyui, which gives me a migraine with the interface. sure it's more versatile, but it ain't easy on the eyes
RB_diff I hear you on that, it does get easier though.
srry to come late but is this compatible with comfy or? if it's not the case should i stick with the facedetailer node in comfy?
It's compatible using Impact pack https://github.com/ltdrdata/ComfyUI-Impact-Pack and Subpack https://github.com/ltdrdata/ComfyUI-Impact-Subpack Custom Nodes. I do not have a workflow handy, but if your face detailer node accepts segm model you should be able to use this with it.
@cranpeach could you please bring me a workflow or i can give you a workflow i have and you manage to update it it's really urgent brother i'll be rewardful im finishing my pipeline right now before i start big biz soon, all ppl helped me i'll count them, telegram @jinbeers
or if you have discord or whaveter.
@carlitajohn89931 Sure, heres an SDXL Workflow: https://civitai.com/images/119604404 and a Z-image Turbo Workflow: https://civitai.com/images/119605093. The model goes in ComfyUI\models\ultralytics folder.
@cranpeach thx, but im wondering if this is more powerful that facedetailer? and which difference it bring, writing this before test
@carlitajohn89931 This is not a custom node, this is a model to use with face detailer to detect only eyes and nose but not the mouth. It is for specific uses where you would not want to detail the rest of the face.
@cranpeach got you!
This works flawlessly in ComfyUI with the Face Detailer node (gotta load it into the "UltralyticsDetectorProvider" node if anyone is wondering).
sometimes it only detects one half of face, how can i set it up to work properly?