MP4 to WebM

WebM is the royalty-free video format made for the web. Every major browser plays it in a page’s video element, and some sites and tools ask for it by name. This page makes a WebM with VP8 picture and Vorbis sound from any video FFmpeg reads — MP4, MOV, MKV, AVI.

VP8 and Vorbis, not VP9 and Opus. The newer pair compresses better, but in this page’s FFmpeg build the VP9 encoder crashes inside the browser, and so does the Opus encoder on anything but mono sound — so they are not offered rather than offered broken. VP8 and Vorbis are the codecs WebM started with, and every browser that plays WebM plays them. The picture is set to constant quality, CRF 10, with a ceiling of 0.1 bit per pixel per frame so a busy scene cannot run away with the file size. Sound that is already Vorbis or Opus is copied; anything else is re-encoded as Vorbis at quality 4, about 128 kbps by Xiph’s own table.

First the page asks the browser itself. Browsers carry the video encoders they use for calls and screen recording — H.264, VP9, AAC, Opus, often on the graphics chip — and when this browser can read your file and write the result with them, that is what happens: nothing is downloaded, the time is measured here first, and the line under the badge says the browser’s encoder did the work. MP4, MOV, WebM and MKV files usually qualify. AVI, WMV and FLV do not, and neither does an MP3, so those go the way below.

The work is done by FFmpeg, the program most video software is built on, compiled to WebAssembly so that it runs inside this page. It is large — 16.8 MB — and it is not downloaded until you press the button that says so. Then it reads your file, encodes 2 seconds from the middle of it as a test, and tells you how long the whole file will take on this device, before anything starts. That figure is measured here rather than taken from a table: the same engine and settings, your processor, your video. It is still an estimate, and it is labelled as one.

It is slower than a program installed on a computer. Here FFmpeg runs on one processor core, because a browser only lets a page use more when the site is served with particular security headers, and it cannot use the graphics chip’s video encoder at all. A long high-resolution video can take many minutes, and longer on a phone. The estimate says so before you commit to it, the progress line shows the time left as it goes, and Cancel stops it at once. When the browser’s own encoder can do the job, it is far quicker, and nothing is downloaded.