When using face after beautification as source image, the generated result seems fail to decouple identity from expression.
Any help about this problem?
May extra face after beautification images finetuning help?
Or the motion extractor from your article should also be funetuning?