@philip-couling's answer is close but missing several pieces:
- The desired size in MB needs to be multiplied by 8 to get bit rate
- For bit rates, 1Mb/s = 1000 kb/s = 1000000 b/s; the multipliers aren't 1024 (there is another prefix, KiB/s, which is 1024 B/s, but ffmpeg doesn't appear to use that)
- Truncating or rounding down a non-integer
length (instead of using it precisely or rounding up) will result in the file size slightly exceeding target size
- The
-b flag specifies the average bit rate: in practice the encoder will let the bit rate jump around a bit, and the result could overshoot your target size. You can however specify a max bit rate by using -maxrate and -bufsize
For my use-case I also wanted a hardcoded audio bitrate and just adjust video accordingly (this seems safer too, I'm not 100% certain ffmpeg's -b flag specifies bitrate for video and audio streams together or not).
Taking all these into account and limiting to strictly 25MB:
file="input.mp4"
target_size_mb=25 # 25MB target size
target_size=$(( $target_size_mb * 1000 * 1000 * 8 )) # target size in bits
length=`ffprobe -v error -show_entries format=duration -of default=noprint_wrappers=1:nokey=1 "$file"`
length_round_up=$(( ${length%.*} + 1 ))
total_bitrate=$(( $target_size / $length_round_up ))
audio_bitrate=$(( 128 * 1000 )) # 128k bit rate
video_bitrate=$(( $total_bitrate - $audio_bitrate ))
ffmpeg -i "$file" -b:v $video_bitrate -maxrate:v $video_bitrate -bufsize:v $(( $target_size / 20 )) -b:a $audio_bitrate "${file}-${target_size_mb}mb.mp4"
Note that when -maxrate and -bufsize limit the max bit rate, by necessity the average bit rate will be lower, so the video will undershoot the target size by as much as 5-10% in my tests (on a 20s video at various target sizes). The value of -bufsize is important, and the calculated value used above (based on target size) is my best guess. Too small and it will vastly lower quality and undershoot target size by like 50%, but too large and I think it could potentially overshoot target size?
To give the encoder more flexibility if you don't have a strict maximum file size, removing -maxrate and -bufsize will result in better quality, but can cause the video to overshoot the target size by 5% in my tests. More info in the docs, where you will see this warning:
Note: Constraining the bitrate might result in low quality output if the video is hard to encode. In most cases (such as storing a file for archival), letting the encoder choose the proper bitrate is the constant quality or CRF-based encoding.
You can use the following code to declare a reusable bash function:
ffmpeg_resize () {
file=$1
target_size_mb=$2 # target size in MB
target_size=$(( $target_size_mb * 1000 * 1000 * 8 )) # target size in bits
length=`ffprobe -v error -show_entries format=duration -of default=noprint_wrappers=1:nokey=1 "$file"`
length_round_up=$(( ${length%.*} + 1 ))
total_bitrate=$(( $target_size / $length_round_up ))
audio_bitrate=$(( 128 * 1000 )) # 128k bit rate
video_bitrate=$(( $total_bitrate - $audio_bitrate ))
ffmpeg -i "$file" -b:v $video_bitrate -maxrate:v $video_bitrate -bufsize:v $(( $target_size / 20 )) -b:a $audio_bitrate "${file}-${target_size_mb}mb.mp4"
}
ffmpeg_resize file1.mp4 25 # resize `file1.mp4` to 25 MB
ffmpeg_resize file2.mp4 64 # resize `file2.mp4` to 64 MB
