How long should a test clip be?

[nghiabeo20]

I want to test several encoding parameters for each film I have, by encoding some sequences with each set of parameter. The goal is medium bitrate (8-10 mbps for 1080). How long should each sequence/clip be (these sequence are randomly selected)? Is 10 second each enough, or I have to manually test the whole cutscene? Thanks!

[StainlessS]

10 seconds is a bit short and could be constant static scene, 20 to 30 seconds would give more chance of having some movement in the clip.
(even then could be a bit static, someone reading a book, or whatever)

[benwaggoner]

What data are you trying to get? If you're trying to estimate average bitrate or something, movies tend to be highly variable in their needs. The end credits may need only 10% the bitrate of an action scene, and a 3:1 ratio between max and typical content required bitrate isn't unusual. I've never seen a sampling scheme that was reliable enough to predict required bitrate that didn't include a full-length analysis. If your constraint is file size, use 2-pass VBR. If your goal is minimum file size for a given visual quality, use CRF. If you need to limit worst-case ABR but save bits on easier content, you should be able to combine --min-crf and --bitrate in a 2-pass VBR encode. Combining CRF and maxrate in a 1-pass encode will cap ABR and PBR. Using 1-pass and bitrate with min-crf tends to starve the start of the movie, which can often be more complex, because the encoder doesn't know it can save bits on the end credits.