Marc's Public Blog - Linux Hacking

After a bit over half a year using it, I figured out that while indeed I don't want to use h265 since it does not decode in every app for licensing issues. At the time, I had also figured out that in ffmpeg, AV1 which is supposed to be even better than VP9, was indeed slower, but not better as far as compression ratio is concerned, or barely enough to matter.

Since then, I came to realize that vp9 is definitely at least 2x slower to encode compare to h264. Originally I considered going back to h264 but then I saw the terrible video quality for lights and lasers, so I'm going to eat the time delay. I will see if for recompressing IG videos before I can merge them, maybe it's fine to use h264 for speed.

vp9 vs av1 comparison and flags:
https://jonathanmh.com/p/encoding-webm-videos-with-ffmpeg-vp9-av1

Gemini says: Compression Efficiency

AV1: Offers the best compression efficiency, resulting in smaller file
  sizes for a given quality level
H.265: Provides significant improvements over H.264, offering better
  compression than VP9 in many cases
VP9: While not as efficient as H.265 or AV1, it still offers decent
  compression and is widely supported.

Encoding and Decoding Speed:
AV1: Requires more computational power for encoding and decoding,
  which can impact performance on older hardware
H.265: Offers a good balance between compression efficiency and
  encoding/decoding speed.
VP9: Generally faster to encode and decode compared to AV1.

# x264 (H.264)	Fastest
# libvpx-vp9	~5–10× slower than x264  => my test shows single pass is 2x slower than h264 2 pass
# libaom-av1	~20–50× slower than x264 (depending on settings)
#
# VP9 (libvpx-vp9): VP9 is a more modern and complex codec. It performs a much
# more exhaustive analysis of the video to find better ways to compress it. This
# allows it to achieve the same visual quality at a 20-50% smaller file size than
# H.264, but this complex analysis requires significantly more CPU time.

# C0012_H264_orig.mp4  (sony camera 1080p h264)
# C0012a_mencoder_H264_4:34.mp4  => terrible quality
# C0012b_ffmpeg_H264_2:08.mp4    => better quality and 2x faster encode
# C0012c_VP9_4:19.webm           => decent enough
# C0012d_AV1_22:03.webm          => tiny bit better?

# h4    9mn  91MB  => join worked
# h4ff 15mn  93MB  => non matching but can read android h265 that mencoder cannot
# vp9  45mn  102MB => join worked
# av1  65mn  93MB  => join worked

Vp9 is around 2x slower than h264 but the output quality is definitely better. Doing single pass over 2 pass still gives decent output. Maybe for IG video stories, which has taken over 24H in the past for just 1 to 2H of video, I'm going to go back to h264 for speed's sake.
Never a dull moment indeed, doing video is really a lot more work than pictures....