This is because of the fact that the source data for this interpolation was not captured at a regularly-timed interval. Twenty-three photos comprise the sequence, and six of those photos have been determined to have gaps of roughly three seconds between them, while others are just a second and a half apart. […] As such, it is extremely difficult, even with the most advanced interpolation methods currently possible, to create a seamless, 100% realistic interpolation of the original photographic sequence. […] The reason this particular interpolation looks rough is simple: It’s based on a set of original source images captured from a standard-definition documentary that aired in 1990. The screenshots taken of the sequence in that documentary, while decent, were insufficient for the methods applied. They were improved using AI sharpening and enhancement methods, yet still did not match the quality and resolution of the original sequence. As such, the interpolation had a lot of missing “data” to fill in. Since this interpolation, a greater-resolution product was produced ( https://www.youtube.com/watch?v=rD-RldBQx7U ) , however even further work is currently in progress.
Damn that looked even worse than I imagined it would. The movement doesn’t even naturally follow one shot to the next.
It explains why in the pinned comment: