Resize a video

Two complaints both get called "too big". A file that a form rejects at 100 MB is a bitrate problem; a video that stutters on an old laptop is a resolution problem. They have different fixes, and the fix for the second happens to be very good at the first as well.

Going from 3840 pixels wide to 1920 removes three quarters of the pixels, and the file follows roughly in proportion. No quality slider gets near that, because the slider is deciding how carefully to store pixels you did not need to keep at all.

Nothing stops you typing a width larger than the source, and the encoder will produce it — a bigger file holding the same information, softened. The detail was never recorded and no amount of interpolation invents it back, so the number to type is the one you actually want the video to be, downwards.

Some tools now offer machine-learning upscaling, which fabricates plausible detail rather than recovering real detail. That is occasionally useful and never faithful, and it is not what somebody trying to fit an upload limit is asking for.

This is one of the operations that genuinely re-encodes: every frame is decoded, scaled and encoded again. On a three-minute clip that is a few seconds; on an hour of 4K it is a serious piece of work for a browser tab, and it is bounded by your machine rather than by us.

The audio is carried across untouched. There is nothing to gain from re-encoding sound when the complaint was about the picture, and doing it anyway would cost a generation of quality for no reason.

Changing width and height independently stretches the picture. A 16:9 video forced into a 4:3 box makes everyone in it noticeably wider, and there is no setting that undoes it afterwards — the pixels have been resampled.

Where a specific frame shape is genuinely required, the correct approach is letterboxing: scale the video to fit inside the target and fill the remainder with black bars. That keeps the geometry honest at the cost of unused edges, which is the trade every broadcaster and platform makes.

Reducing a video’s dimensions averages several source pixels into each output pixel, which suppresses noise and compression artefacts. A 4K clip at 1080p frequently looks cleaner than a clip shot at 1080p, because the detail was oversampled.

The reverse is not true in any useful sense. Enlarging invents pixels from the ones nearby, so a 480p video at 1080p is a soft 480p video occupying four times the space. Nothing on this site or anywhere else recovers detail that was never recorded.

H.264 and H.265 work on blocks and store colour at half resolution, which means both width and height must be even numbers. An odd dimension is rejected by the encoder rather than silently rounded.

This surfaces when scaling by a percentage — 55 per cent of 1080 is 594 and 55 per cent of 1921 is not an integer at all. Choosing a standard height and letting the width follow, or rounding both to a multiple of two, avoids an error message that does not explain itself.

A smaller frame needs fewer bits to look the same. Halving the dimensions quarters the pixel count, and keeping the original bitrate simply spends it on encoding the same picture more precisely than anyone will see.

If the point of resizing was a smaller file, that is where the saving actually comes from — the dimension change alone does much less than expected if the encoder is told to hold the old bitrate. Reducing both together is what turns a resize into a meaningfully smaller video.