I haven’t tried HEVC in a couple of years, so I did some fresh testing to see if my previous reasons were still valid.
I found H.264 CRF 17-18 Medium to correspond to HEVC CRF 20 Slow (Shotcut quality 60%). But to my eye, these CRFs are “good enough for delivery” and not “visually lossless for archive”. Below are the general purpose settings I consider to be visually lossless enough for archival purposes and also able to survive a generation of transcoding (note this is totally subjective to my own video material):
- H.264 CRF 16 (Shotcut quality 68%), preset Medium
- H.265 CRF 18 (Shotcut quality 64%), preset Slow
Yup, the Slow preset was required for H.265. I tried Medium first and the fine details were always smeared. Medium never looked quite right even up to very high-quality CRFs. The slow preset is a substantial jump in quality.
For extra credit, I would consider using QP instead of CRF for any intermediate files to avoid the bitrate shortcuts that CRF uses during fast motion sequences. QP would eliminate practically all chances of macroblocking.
Now for the trade-off. Using the settings above, H.264 makes a file that is 2x the size of HEVC. However, HEVC takes 6x the time to encode the file as H.264.
So the question is processing time versus disk space. Since we have slow hardware, processing time is our biggest concern. If we have an export that takes two hours with H.264, it would take 12 hours with HEVC. That is enough time to either delay a YouTube video release by a day, or to cut a day out of the post-production schedule in order to leave time for the export (and no time for a do-over if there was a mistake). That’s brutal on the production timeline. Plus, we can’t edit the next video during the additional ten hours that HEVC requires, so that’s a double penalty.
If we had faster hardware or a GPU that could achieve similar quality in similar time, then maybe we could justify HEVC. But for now, time is more important than disk space when the space difference is only 2x versus the time penalty.
After my tests, I found a well-researched article that reached the same conclusions I did in terms of quality settings:
So I guess these settings should be pretty reliable for anyone else that wants to use them.