XAVC-I v ProResHQ. Just how good is XAVC-I?

Описание: I often hear people saying that XAVC-I isn't good enough or that you MUST use ProRes or some other codec for the best image quality. My own experience is that XAVC-I is actually a really good codec and recording to ProResHQ only ever makes the very tiniest (if any) difference to the finished production. I've tested this many time in the 8 years or so that I have been using XAVC-I but to illustrate my own findings I decided to put together this (rather boring) video.

It was shot with a Sony FX6 using internal XAVC-I (class 300) on an SD card along side an external recording using ProResHQ on a Shogun 7.

I deliberately chose to use Cine EI and S-Log3 at the cameras high base ISO of 12,800 as noise will stress any codec that little bit harder. Making this test just that little bit tougher. The slightly higher noise level also allows you to see how each codec handles noise more easily.

A sample clip of each codec was place in the time line (DaVinci Resolve) and a caption added. This was then rendered out, he ProRes HQ filess rendered using ProRes HQ and the XAVC-I files rendered to XAVC-I. So for most of the examples seen the XAVC-I files have been copied and re-encoded 5 times plus the encoding to the file uploaded to YouTube, plus YouTubes own encoding, a pretty tough test.

Because in most workflows I don't believe many people will use XAVC-I in post production as an intermediate codec I also repeated the tests with the XAVC-I rendered to ProRes HQ 5 times as this is probably more representative of a typical real world workflow. These examples are shown at the end.

Of course the YouTube compression will restrict your ability to see some of the differences between the two codecs. But, this is how many people will be distributing their content. Even if not via YouTube via other highly compressed means, so it's not an unfair test. Looking at the original clips in my studio it is very hard to spot the differences.

Where the s709 LUT has been added it was added AFTER each further copy of the clip, so this is really a "worst case scenario", the LUT being added to the lowest quality S-Log3.

Overall in the end the ProResHQ and XAVC-I are remarkably similar in performance. In the 300% blow up you can see differences between the XAVC-I that is 6 generations old compared to the 6th generation ProResHQ. But the differences are very, very hard to spot and going 6 generations of XAVC-I is unlikely in most normal workflows. In the same test where the XAVC was rendered to ProResHQ for each post production generation any difference is incredibly hard to find even when magnified 300%.

I am not claiming that XAVC-I Class 300 is as good as ProResHQ, but I think it is worth considering what you need when shooting. Do you really want to have to use an external recorder, do you really want to have to deal with files that are 3 to 4 times larger. Do you want to have to remember to switch recording methods between slow motion and normal speeds? For most productions I very much doubt that the end viewer will ever be able to tell the difference between material shot using XAVC-I class 300 and ProResHQ, even if watching with their face inches from a very large screen.