Containers, codecs, and when a conversion costs nothing
The extension names the box, not the contents. An .mp4, .mkv, .mov and .webm are containers: they hold tracks, timestamps and some bookkeeping. What the picture actually is — and what every player has to be able to decode — is the codec inside it: H.264, HEVC, VP9 or AV1 on modern files, DivX, Xvid, Sorenson Spark or VP6 on old ones. Two files with the same extension can be entirely different jobs.
That is why "convert to MP4" sometimes takes seconds and sometimes takes an hour. When the codec inside can live in an MP4, the encoded video is copied across byte for byte and only the box is rebuilt: a two-hour H.264 MKV or MOV converts in seconds and the picture is identical, because nothing was decoded. When the codec cannot live in an MP4, or can but almost nothing plays it, the video has to be decoded and encoded again — DivX and Xvid in an AVI, VP8 and VP9 in a WEBM, Sorenson Spark and VP6 in an old FLV. That is a real re-encode, and it costs time and a little quality. Each tool says which path it took.
Every re-encode is a fresh lossy pass, and they stack. A clip that has been through three converters has been squeezed three times, and no setting recovers what the earlier passes discarded. Where you have the choice, do the whole job in one step rather than converting, then compressing, then converting again.
Trimming is only frame-accurate if it re-encodes, and that is how compressed video works rather than a shortcoming of the tool. Most frames describe only what changed since the frame before, so a copied cut can begin only on a key frame — a second or two apart in most files. Trim Video tells you the time it actually used, and switching on the exact cut re-encodes the start so it lands where you asked. The same arithmetic is why Rotate Video is instant on MP4 and MOV: those formats carry a rotation flag, so turning a clip rewrites a few bytes and touches no pixels. AVI, WMV and MPEG have no such flag, so the turn is baked into the picture and the file comes back as an MP4.
MP4 is a poorer box than MKV, and that is where files lose things. MKV was built to hold many tracks: several languages, subtitles, chapters, attached fonts and cover art. An MP4 that everything plays holds one video track and one sound track. So a converted file keeps the main picture and sound and leaves the rest behind, and you are told what was dropped rather than finding out later. Dolby Digital and DTS sound is the other common casualty: it is converted to AAC here, because the alternative is the file that plays silently on a TV.
Where a browser is genuinely the wrong tool
These tools encode through WebCodecs, which hands the work to the video encoder built into your device — the same silicon a video call uses. That makes them far faster than the WebAssembly ffmpeg builds most browser tools run on, and it is where the good news stops. We would rather set the limits out here than have you find them halfway through a job.
Only H.264 and AAC come out. Nothing here encodes HEVC, AV1 or VP9, because browsers do not offer those encoders. H.264 is the most compatible codec there is and the least efficient of the modern ones, so anything re-encoded here is larger than the same quality from x265 or SVT-AV1. If file size is what you are optimising, HandBrake and ffmpeg are both free and will beat this comfortably.
There are three compression settings, not a number. Light, Balanced and Strong, with an optional drop to 1080p or 720p. No CRF value to type, no two-pass encoding, no encoder tuning, no way to target an exact file size. And there are no filters at all: no deinterlacing, so an interlaced AVI keeps its combing; no denoise, no colour grading, no cropping, no burned-in subtitles, no arbitrary rotation angle, and no crossfades or titles when merging. HDR is converted to standard range, so a bright HDR clip comes out flatter than it went in.
Your device sets the pace and the ceiling. Output is written to the browser's own storage as it is produced rather than held in memory, so length is limited mostly by free disk space — but a real re-encode runs several times faster than real time on a modern laptop and around real time on a phone, and a phone will get hot and give up on a long 4K file. A copy — a rewrap, a mute, a plain trim, a flag rotation — is seconds whatever the length, because nothing is decoded.
And it runs when a person clicks, on one file. Converting four hundred clips on a schedule is a shell loop around ffmpeg, which is free and better at it than any web page will ever be.
Choosing between the tools that sound alike
MP4 to MP3 vs Extract Audio. They start the same way and end differently. MP4 to MP3 always gives you an MP3, at a bitrate you choose. Extract Audio lets you pick MP3, M4A or WAV — or keep the sound exactly as the video stored it, copied across without re-encoding, which is a perfect extract and takes seconds. Take Extract Audio if you are going to edit the sound; take MP4 to MP3 if you want something that plays in a car.
Extract Audio vs Mute Video. Extract Audio takes the sound out as its own file and leaves the video alone. Mute Video removes the sound track from the video and leaves you the picture. They are two halves of the same split, and running both gives you both — losslessly, because neither re-encodes a modern file.
MKV to MP4 vs Compress Video. MKV to MP4 changes the box and leaves the file roughly the size it was; it is not a way to make a film smaller. Compress Video changes the bits, which is the only thing that does make a file smaller, and it re-encodes to do it. If the problem is that a TV will not play the file, convert. If the problem is that it will not fit in an email, compress — and scale it down while you are there, because halving the height saves far more than any quality setting.
Trim Video vs Video to GIF. Both take a stretch out of a clip. Trim gives you a video, usually in the format you started with and usually without re-encoding. Video to GIF gives you a silent looping image in a format from 1989: 256 colours a frame, every frame stored in full, no motion compression at all — which is why three seconds at 480 pixels wide weighs a megabyte or more. Use it for a chat or an issue tracker, and use Trim for anything longer than about ten seconds.