Datamosh
No filter here. Both clips are really compressed, like any video, and then the second one's keyframe is removed. The decoder gets B's motion but not its picture, so it moves A's pixels instead. How it works inside, in the guide →
Clip A · the picture that stays
Clip B · the motion
B's keyframe is removed: its motion drags A's picture along.
With lower quality the video leans more on motion and the datamosh gets wilder.