Understanding video cropping
Pick a rectangle, drop the rest.
What cropping a video means for the codec, the difference between crop and scale, and the aspect-ratio decisions that come with every reshape.
Crop is reframing.
A crop operation discards everything outside a chosen rectangle. The remaining pixels keep their resolution; the output's dimensions become the rectangle's dimensions. A 1920×1080 video cropped to a 1080×1080 square in the centre becomes a 1080×1080 video; the cropped sides are gone. Cropping always involves re-encoding (the codec can't represent "the same data with a different frame size") — there's no stream-copy fast path.
Crop vs scale.
Crop discards pixels; scale resamples them. Cropping a 1920×1080 video to 1080×1080 keeps the centre pixels at their original resolution and drops the sides. Scaling the same source to 1080×1080 squashes the whole image into the target — same content, distorted aspect ratio. For platforms that require a specific aspect ratio (Instagram 1:1, TikTok 9:16), crop is usually the right answer; scale produces visible distortion.
A worked crop.
Crop a 1920×1080 video to 1080×1080 centred on the middle. ffmpeg -i in.mp4 -vf "crop=1080:1080:420:0" -c:a copy out.mp4. The crop filter takes w:h:x:y — output width 1080, output height 1080, x-offset 420 (centres horizontally on a 1920-wide source: (1920-1080)/2 = 420), y-offset 0. The audio stream-copies; the video re-encodes with the cropped dimensions.
Centre crop to square
crop=1080:1080:420:0
Width:height:x-offset:y-offset, all in pixels.
(1920-1080)/2 = 420 horizontal offset
= 1080×1080 square output
Vertical (9:16) for shorts
crop=ih*9/16:ih
FFmpeg expressions: ih = input height, scale width accordingly.
From 1920×1080 → 607×1080 (vertical strip from centre)
= Phone-ready vertical video
Picking the right rectangle.
For an interview shot (one person centred), a tight centre crop works. For a two-person frame, centre-cropping cuts both heads off; you'd want a wider crop or a different aspect ratio. For motion content (sports, action), what's in the centre might not be the subject — you'd need a tracked crop. The simple fixed-rectangle crop is right for static framings only; for moving subjects, more sophisticated tools (AI-driven content-aware crops) exist but cost more compute.
Even dimensions matter.
H.264 and H.265 require even-numbered dimensions (the chroma subsampling at YUV 4:2:0 averages pairs of pixels in both directions). Cropping to 1081×1081 fails; the encoder rejects the dimensions. Either round to even (1080×1080) or usecrop=...,scale=... to massage the dimensions before encode. Most tools handle this automatically; raw FFmpeg commands occasionally trip over it.
Re-encoding is the cost.
Every crop is one generation of lossy re-encode. For a clip you'll edit further, accept the cost once. For an archive copy you'll just view later, sometimes letterboxing the playback (instead of cropping the source) is the right call — keep the original intact, display it in a smaller window with bars. Decide based on whether the output is the canonical version or a derivative.