Video Converter

Container and codec: the difference that decides everything

A video file is two things at once. The container — MP4, MOV, MKV, WebM, AVI — is the box: it holds the streams, their timing and their metadata. The codecs — H.264, HEVC, VP9, AAC, Opus — are how the pictures and sound inside the box are compressed. 'Convert MKV to MP4' is a request to change the box; whether the contents also have to change depends on what is in them.

When the codecs inside the source are ones the new container supports, and that the players opening it expect, the streams can be copied across byte for byte. This is called remuxing. It takes about as long as reading the file, involves no decoding, and loses nothing: the output looks and sounds identical to the input.

When they are not — VP9 video heading into an MP4 meant for an iPhone, H.264 heading into WebM — the stream has to be decoded and encoded again. That is the slow part of any conversion, and it costs a little quality each time, which is why this converter never does it unless it must.

What each output format is for

MP4 is the default and the right answer for almost everyone: it is what phones, browsers, social sites, PowerPoint and editing software all accept. Here it means H.264 or HEVC video with AAC or MP3 audio. H.264 and HEVC sources are copied; VP8, VP9, AV1, MPEG-4 Part 2 (Xvid/DivX), WMV and others are re-encoded to H.264.

WebM is the open web format. Output is VP8 video with Opus audio; VP8, VP9 and AV1 sources are copied. VP9 encoding is not offered, deliberately: the VP9 encoder in this browser build of ffmpeg crashes on the first frame, and offering a broken option would be worse than leaving it out.

MOV is QuickTime, for Final Cut Pro, iMovie and other Apple tools; it additionally keeps ProRes and Apple Lossless audio. MKV accepts almost any codec, so nearly everything is copied — useful when you simply need a file a particular player will open, without touching the streams at all.

Options, and what each one costs

Method: Auto copies everything it can. Always re-encode forces H.264 (or VP8 for WebM) even when copying would work — the choice for a site or device that rejects HEVC, or for a source with broken timestamps that plays badly when copied.

Audio track: when a file has several, choose one. MP4 can carry more than one, but most phones, browsers and TVs only ever play the first, so one well-chosen track is the reliable option. You can also drop the audio entirely.

Resolution: 1080p, 720p or 480p, only ever scaling down. Any resize forces a re-encode, so it is the right control when the goal is a smaller file as well as a different format; for size alone, the Video Compressor gives finer control.

Quality applies only when re-encoding. Balanced uses x264's veryfast preset at CRF 23, a good default; Higher quality uses CRF 20 for a closer match and a bigger file; Fastest uses the ultrafast preset, about two and a half to three times quicker but noticeably larger.

How long it takes

A remux runs at the speed the browser can read the file — typically several hundred megabytes in a few seconds. That covers most iPhone MOVs, most OBS and archive MKVs, and most MP4s going to MKV or MOV.

A re-encode is real work, and it runs single-threaded. Measured on this build on a recent desktop processor, H.264 at the default setting manages about 3–4 frames a second at 1080p and 8–9 at 720p: roughly eight minutes per minute of 30 fps 1080p footage, three to four for 720p, and around a third of that with Fastest. Laptops and phones are slower. Keep the tab in the foreground — browsers throttle background tabs heavily.

Single-threaded is a deliberate choice. The multithreaded build of ffmpeg needs the page to be cross-origin isolated, and that isolation would break other parts of the site; the speed trade is the price of keeping everything in the browser.

What is not carried over

Subtitles are not copied: MKV's SRT and ASS subtitle tracks do not fit MP4 in their original form. Extract them with a desktop tool if you need them, and use the Subtitle Converter to convert or re-time them.

Only the selected audio track is kept, and only the main video stream — cover art and secondary video tracks such as depth maps are left out. Global metadata such as the title and creation date is copied. A portrait phone video keeps its rotation flag when copied and is physically rotated when re-encoded, so it plays upright either way.