Any way to convert not using a model but with provided faces?
I was thinking of using the First Order Motion Model by Aliaksandr Siarohin with the Vox checkpoint to animate an image but it requires cropping the target video to be roughly centered on the face. Manually doing that seems like a pain for anything with varying camera angles so it would be really convenient if I could use FaceSwap to generate the cropped faces and just feed that in. Is there some way I can use the alignment file to then just take the swapped faces generated by the First Order Motion Model and swap them back into the target video?