Open NeoAnthropocene opened 2 weeks ago
in paper, the authors mentioned that they use some "human face recognition" network to extract the face's feature and embed original ID face's feature during training, so it is why it won't work for animal, object or anime. I wonder can we avoid this issue by using some "everything-recognition" network
β Just a heads-up for other creators like me who want to use the PuLID in their workflows: it doesn't support anime and animal faces. InstantID can detect at least the anime character faces.
β It seems like the EVA CLIP is already handling the face detection. I'm curious though if there's a way to make it even better at face detection like the InstantID method does.
π Reference to the images that fails on PuLID face detection: