Our Goal is to implement ZAO Faceswap Effect perfectly with same performance(translated only by one image, less than 5 seconds each process) and effect(not weired).
we have tried following models ,and these methods failed due to performance and effect.
really fast, but it is too fake!
some methods need to be implemented