The 2 problems with deepfakes(at least with the for-the-public versions) are that they haven't figured out how to do the hair and also that they are mapping the fake onto the facial/head outline of the original person which can be very different sometimes. For an extreme example, if the initial person had an entirely spherical face, they'd have to map Phil Heath's features onto a sphere. You see the perimeter outline of Schwarzenegger's actual face, like the edge of a silhouette, from any 2-D viewing angle at all time. If a feature falls within that perimeter, like the nose or lips in many angles, you can morph those features to appear outside the original subject's 'geometry' and have a bigger nose, lips, eyes, etc.... Phil's enlarged head is restricted by the outline of Schwarzenegger's head from any angle. It works best when head shapes, features, and hair are similar.
