Overall, not bad. However, check out that aliasing and sharpening on the titles!


Some of the matte shots looked de-grained, but they were few.
Thoroughly Modern Millie (1967) was processed by Universal.
Taking a guess, it was de-grained, restored to death, then re-grained. It looks smeary and unnatural.
And check out those ugly grid lines!
Interpolation still shows under the layer of false grain.


Now, compare to Autour de la Fin du Monde (1930). There’s a clarity that comes through the softness when the picture is left alone:
Too bad it’s out of sync:
La fin du monde (1931) was processed by Gaumont, Eclair, and FPA France. This version is also available from Kino.
Very nice video. No complaints.
However, it’s over-subtitled and the translation is lacking.

We don’t need the entire Lord’s Prayer subtitled. It’s a pretty standard execution scene. Just this line, ending with an ellipsis, is enough.

The Spanish word for “no” is “no.” Do not subtitle cognates.

The title character’s nickname (“El Gato”) should be “The Cat” in English. (Though, in this example, the cop simply calls him, “Gato.”) We say “Mack the Knife,” not “Mack Messer.”
Coloring the subtitles pale yellow is a nice touch. I approve.
The Murnau Foundation put a ton of work into piecing the film back together. Then they had to go and blow two years of work by “restoring” it.
Carmen was reconstructed from multiple prints. Many, many gaps were filled in from lesser sources, often no more than a few frames. However, the technicians were a little sloppy, repeating frames they already had, resulting in added stutter, microloops, and reduced picture quality.


In this sample, the technicians appended three frames to the end of the shot. Two of the frames were already present, resulting in a microloop:
Criterion transferred The Unknown at a little under 23fps and encoded at 1080i. The motion should be perfect, right? Wrong! Somebody messed up and used a 24fps timeline, then exported at 1080i60. Notice how the motion stutters in the top sample below. The whole reason to use 1080i is to prevent that!
After deinterlacing, we can easily see the uneven, broken pulldown, worse than most silents authored at 24fps. Below screenshot is of a tool I wrote to help delete duplicated frames. What you see on the right is a long list of frame numbers. All frames highlighted in red are duplicates and can be safely deleted.

Encoding silents to 1080i60 applies its own unusual pulldown. So, the result of working in 24fps and exporting at 1080i60 is that two different pulldown patterns compounded to create terrible stuttering.


Frozen cloning doesn’t match:
