How to compress a video (without guessing at bitrates)

Constant quality, not a bitrate you guessed

Every preset here is built around CRF — constant rate factor. Instead of telling the encoder "use 4 Mbps", you tell it "keep this level of quality" and it spends whatever bits each scene needs. A still shot of a wall costs almost nothing; a handheld pan through leaves costs a lot. A fixed bitrate wastes bits on the first and starves the second, which is why fixed-bitrate files so often look fine for ten seconds and then fall apart.

CRF runs backwards from intuition: lower numbers mean higher quality and bigger files. Our quality presets sit at CRF 18 for Ultra, 21 for High, 23 for Balanced and 30 for Maximum compression. The gap between 18 and 23 is much smaller to the eye than it is on disk — that is the whole reason Balanced is the default.

Everything is encoded to H.264 video and AAC audio in an MP4 container, with the index moved to the front of the file so it starts playing before it has finished downloading. That combination is the one that plays everywhere without a conversation about codecs: phones, browsers, TVs, messaging apps, and every editor you might drop it into next.

Let the destination pick the setting

The quality presets are the right answer when the video is for you — an archive copy, a file to edit later, something going onto a drive. The platform presets are the right answer when it is going somewhere specific, because that somewhere has rules, and matching them is not a matter of taste.

WhatsApp is the clearest case. Its preset caps the video at 720p and 30 fps at CRF 26, and it carries a hard ceiling of 16 MB on the finished file. Discord is tighter still: 720p at CRF 28 under a 20 MB ceiling. Email sits at CRF 27 under 24 MB. Those ceilings are not advice — when one is set, the encoder is told to fit under it, and the result comes back under it.

The looser presets do less because less is needed. Instagram and Facebook cap at 1080p and 30 fps at CRF 23. YouTube deliberately does not cap resolution at all and encodes at CRF 20 with 192 kbps audio, because YouTube is going to re-encode whatever you send: your job is to hand it the cleanest possible source, not a small file.

If you are not sure, the honest default is Balanced. It caps at 1080p and CRF 23, which is the setting most people would have arrived at after an afternoon of experiments.

Resolution is the other half of the saving

Halving the height of a frame quarters the number of pixels in it. That is why the aggressive presets cap resolution rather than only raising CRF: dropping a 4K phone recording to 720p removes about nine tenths of the pixels before the encoder makes a single quality decision, and on a phone screen almost nobody can tell.

The caps only ever cap. A video that is already 720p and sent through a preset that allows 1080p stays at 720p — nothing is ever enlarged. Enlarging would invent pixels that were never recorded, producing a bigger, softer file, and a tool that did that quietly would be misrepresenting what it produced.

Frame rate works the same way. The presets aimed at messaging and email cap at 30 fps, so a 60 fps clip loses half its frames and roughly a third of its size. For a screen recording or a talking head this is invisible. For fast sport or gaming footage it is not, which is why the quality presets and YouTube leave frame rate alone.

You see the estimate before you upload anything

When you pick a file, the workspace reads it in your browser and shows what it found — duration, resolution, size — and an estimate of the result for the preset you have selected. Change the preset and the estimate changes. None of that costs an upload: the file has not left your device at that point.

This is deliberately the same calculation the worker will run, not a separate guess written for the interface. A preview that used different arithmetic from the encoder would eventually promise something the encoder does not deliver, and the number on screen would quietly become decoration.

Once you start the job the file uploads and processing begins, with real progress reported as it happens rather than an animation that fills up regardless. When it finishes you get the actual output size and the actual percentage saved, measured on the file you are about to download.

What comes back, and what happens to the original

You get an MP4 with no watermark, at the size the job actually produced. There is no signup, no account and no paid tier hiding the good settings — every preset described here is available to everyone, including the ones that do the most work.

You can send up to 2 GB per file and ten files per job, which covers a phone recording of almost any length. Both the file you uploaded and the file we produced are deleted automatically 24 hours later; the download link stops working at the same time, so save the result before you close the tab for the day.

Nothing about the original is altered — compression happens on a copy. If the result is not what you wanted, change the preset and run it again on the same file rather than re-compressing the output. Compressing an already-compressed video stacks the losses: each pass throws away detail relative to what it was given, so two passes at Balanced look worse than one pass at Maximum compression while saving less.