A light wrapper around serengil/deepface to do facial recognition in Comfy.
Want to make your workflow cleaner? UE Nodes broadcast data without the wires.
Pause the workflow while you pick an image? Image Picker.
You need to install deepface, and the default library used is facenet
pip install deepface
pip install facenet-pytorch
Other possible dependencies - don't install these unless you're told to!
Then install the custom node:
cd [comfy]/custom_nodes
git clone https://github.com/chrisgoringe/cg-faces
Deepface uses various underlying models (check their page for details) which will download on first use.
The nodes all take a single image of the face you are comparing with, and one or more candidate images. Each candidate is compared with the true_image, and given a score for the similarity of the face (just the face - nothing else is taken into account). An image which doesn't have a face in scores 0 (if there are multiple faces in the candidate image the best match is used) a perfect match scores 1. Above 0.6 is generally considered a match.
Nodes have a message output which is a string, either giving some similarity numbers or an error message. Install display text from quicknodes if you don't have a node to display a text message.
Nodes either give a score, or sort the images, or return those above a threshold... usage should be pretty obvious.
