cat

ffmpeg -i "$1"    \
  -stats_period 600 \
  -c:a copy       \
  -c:s copy       \
  -c:v libsvtav1  \
  -crf 23         \
  -preset 3       \
  -tune 0         \
  -svtav1-params "tune=0" \
  "reencoded_$1"

I was thinking of just splitting out the video stream and running it through the svt-av1 encoder directly though since it has some neat other options that I'm struggling to poke at through the ffmpeg layer

I still gotta review the output of this run, but it took the 7.1 GB dvd video and smaashed it down to about 2.2 GB with those settings in the case that just finished. scrubbing over it initially, the main feature looks fine, but some of the special featurettes I'm afraid I might need to try again with some deinterlace/dcomb jazz since they're looking a lil weird

Plum @plumpan5/2/2024, 6:08 AM

ohhhhhhh interesting, how was the encode speed on av1?

Is the input interlaced? -vf yadif=1 should sort it

cat @catball5/2/2024, 6:57 AM

The encode speed is def slower on AV1 in my experiences so far! but it also seems like it saves a lot of space compared to h265. if you're curious about benchmarks of different presets/tunes/params, someone did a giant benchmark comparison with some animated clips here: https://wiki.x266.mov/blog/svt-av1-deep-dive

I checked a bit ago and yeah, the video I encoded (ran 1985 criterion dvd rip) was definitely interlaced :( I'm re-running with yadif (and also trying with quantization matricies enabled):

ffmpeg -i "$1"    \
  -nostats        \
  -c:a copy       \
  -c:s copy       \
  -c:v libsvtav1  \
  -vf yadif=1     \
  -crf 23         \
  -preset 3       \
  -tune 0         \
  -svtav1-params "tune=0:enable-qm=1" \
  "reencoded_$1"

def open to any advice you have! also if you're curious to peek at the results & logs, lmk and I'll chuck them somewhere on the fileserver