FFmpeg Cheatsheet - Video, Audio & Format Conversion

The canonical form on this sheet is ffmpeg -i input.mp4 -hls_time 10 -hls_list_size 0 output.m3u8. -hls_time 10 cuts a segment every 10 seconds; keep that value aligned with the keyframe interval of your encoded output, because cuts only join seamlessly when they land on keyframes, and too short a segment bloats the index. -hls_list_size 0 keeps every segment in the playlist, which is what you want for a catch-up style VOD list. For live streaming with a sliding window instead, pass a positive number — 6 keeps only the six most recent segments.

Use ffmpeg -i http://example.com/stream.m3u8 -c copy output.mp4. The -c copy is the important part: the segments behind the playlist are already in a playable encoding, so stream copy avoids re-encoding, runs at close to network speed and loses nothing in quality. The command has to run start to finish — ffmpeg follows the playlist as it grows and only finalises the file when the stream ends. If the resulting container is wrong, do a separate -c copy remux afterwards rather than changing codecs here.

For a fixed rate use -r: ffmpeg -i input.mp4 -r 1 frame_%03d.png writes one image per second, which suits thumbnail grids and list-page covers. For keyframes use a filter using select=eq(pict_type\,I) with -vsync vfr, which emits only I-frames, so the count follows the source keyframe density — usually far fewer images, each of them independently decodable. Reach for -r when you want predictable, small thumbnails; reach for select with -vsync vfr when you want semantically meaningful cuts at scene changes.

They govern different things. -crf is constant quality: the encoder decides per-frame how many bits the picture deserves, as in ffmpeg -i input.mp4 -c:v libx264 -crf 23 output.mp4, the default. Crank it up and the file shrinks markedly; run it twice on the same source and you get the same result, which makes it the right choice for archival masters and anything you will edit later. -b:v is constant bitrate — you name a number and the encoder hits it, as in ffmpeg -i input.mp4 -c:v libx264 -b:v 1M output.mp4, which suits delivery targets with hard bitrate ceilings and decoders that want to predict bandwidth. When you need both predictability and quality, compose them: encode once with -crf 18 -preset slow, then cap the finished file at the delivery bitrate.

In ffmpeg -i "image_%03d.png" -r 30 video.mp4 the -r 30 is an output frame rate: ffmpeg reads the image sequence as an unbounded stream of frames and lays it out at 30fps, so total duration is simply image count divided by 30 — forty pictures become a clip barely over a second long. On the input side, %03d is a zero-padded three-digit index, which means the files must be numbered contiguously; a gap or a sequence that starts elsewhere makes ffmpeg stop partway through.