If it's an actor wearing Trump's face like in your typical deepfake productions, you'd expect that.

But the face is nearly identical between the two frames. This is crazy.

I'm no machine learning expert, but I'm high-level familiar with how it works. In a network without memory (like you'd expect with a deepfake) there isn't going to be any feedback path for the network to know that it's overused a single training datapoint. I could very well be wrong but this looks like it could be fabricated to me.

The kicker would be if there were other nearly identical frame pairs throughout the video.