16 tokens & Face Control Pose

[a3nima]

Hey, 3 questions:

• can the embedding model loader be run with 16 tokens as well? it gives an error if I try
• how is the extract face pose image node to be used, will the model try to recreate the facepose from this image?
• is it possible to use more than 1 reference image?

[EndlessSora]

Thanks for your interest! Here are some answers for your questions:

1. The embedding model loader only supports 8 tokens. That is what the image project model outputs and the setting we trained on.
2. Yes. The extract face pose image node will extract five-facial-keypoint image given the input image containing a face. The face on the final generation will be controlled by the five-facial-keypoint image.
3. If "reference image" refers to the ID image, the current InfiniteYou (InfU) does not support multi-ID injection.