Right, I looked at that too and dismissed it, but then saw a gif comparing the two frames and the body/background is on a different layer than the head/face.

It’s possible /pol is hoaxing us, but if that repeated frame of Trump’s face, with the exact same compression artifacts, is baked into the actual original video posted, then someone has manipulated it. That could not occur in camera naturally.

This got me thinking more and that is absolutely the way a video like this would be composited. You would pull the body and background from your source and just use a continuous stretch where his hands didn’t move much or moved when you wanted them to. And you could slide that around to be positioned in time wherever you wanted. The head/face would be the hard part and would need to be attached to the body later. You could not use whole frames and chop up the body every time you needed to match a syllable. The shoulders would be jumping all over the place. The body would need to be one continuous layer without cuts. This would both look the most realistic and be the fastest/easiest approach.