Posting my soundpost workflow for youtube sources since apparently everyone else is way less autistic
Find a rough part of a video and feed yt_dlp
>python -m yt_dlp https://www.youtube.com/watch?v=uBGgYVJ1ZOk --download-sections="*27:40-28:00"Then one or more of the following:
Open it up, figure out more precise timestamps (-ss start -t duration)
Take a screenshot to measure crop in GIMP or whatever else (-vf crop=width:height:x:y)
Determine the target video bitrate to stay under 4 MB, roughly 32000k/duration, usually set it lower for some wiggle room (-b:v bitrate)
>ffmpeg -i "【Q REMASTERED】good brain, good squish [uBGgYVJ1ZOk].mkv" -c:a copy -ss 18.8 -t 1.5 -vf crop=540:800:1380:280 -b:v 21333k name.webmThen strip out the components
>ffmpeg -i name.webm -an -c:v copy namevideo.webm>ffmpeg -i name.webm -vn -c:a copy nameaudio.opusVideo is reencoded once, audio is never reencoded. Result picrel
