It does not work well for pictures/video where a target is talking, mostly because mouth shape is not recognized accurately
so results show some average position comparing to the original, no difference between closed and half closed mouth (for example).
For comparison, analogs like SimSwap/simswap-inference-pytorch give very good results in such scenarios but worse in face similarity
It does not work well for pictures/video where a target is talking, mostly because mouth shape is not recognized accurately so results show some average position comparing to the original, no difference between closed and half closed mouth (for example).
For comparison, analogs like SimSwap/simswap-inference-pytorch give very good results in such scenarios but worse in face similarity