Inside a Video File

The container is not the video

MP4, WEBM and MKV are containers — boxes with a defined way of arranging what is inside. The extension names the box and says nothing reliable about the contents.

Which is why ‘it is an MP4, it should work’ is a reasonable expectation that is sometimes wrong. An MP4 usually holds H.264, and it can hold AV1, which older devices cannot decode — containers compared.

Because the box is separate from the contents, changing it is cheap. Repackaging a file from WEBM into MP4 copies the tracks across untouched — no decoding, no quality loss, seconds rather than minutes. That operation is called remuxing and it is what a good downloader does when it hands you an MP4.

Converting the contents is the expensive one. That is re-encoding, it costs a generation of quality, and the two get confused constantly — which is where quality claims come from.

Tracks: picture and sound are separate inside

A video file does not contain video with the sound somehow attached to it. It contains a picture track and a sound track, interleaved through the file, each encoded with its own codec and supported independently.

That separation explains a family of symptoms:

Picture but no sound. The audio track is missing, or in a codec the player cannot decode. Common when a tool hands over a video stream without joining the audio — why that happens.

Sound but no picture. The reverse: the player decodes the audio codec and not the video one, which shows as a black screen — the diagnosis.

A player listing two tracks. Entirely normal. Merging streams into one file keeps them as distinct tracks; it does not fuse them into one thing.

It also explains why extracting audio is quick and lossless when done properly. The sound track already exists as its own thing inside the file — copying it out is a extraction rather than a conversion, and nothing needs decoding.

The same structure allows more than two tracks. A file can carry several audio tracks for different languages, and players offer them as a menu. Platform video almost never does, because the upload processing reduces everything to one — which is why alternative language audio disappears when a video is uploaded.

The index, and why truncation behaves oddly

Somewhere in the file is a table describing what is where — which byte starts which frame, how long the whole thing runs, what the tracks are. In MP4 it has a name, and it is the thing error messages complain about when a file will not open.

Its position decides how a partial file behaves. Written at the end, a truncated download has no index at all and refuses to open. Written at the front — the arrangement used for streaming — the file opens, reports the full duration, and stops early when the data runs out.

That second case is the dangerous one, because the file looks complete. We tested both: the same video cut to a third produced ‘moov atom not found’ in one arrangement and a confident 30-second duration in the other — the full test.

It also explains what repair tools actually do. They rebuild an index so a file opens; they do not restore missing data, and the video still stops where the download did.

What else rides along

• Timing information. Frame rate, and whether it is constant or varies through the file. Variable rates cause audio drift in editors.
• Dimensions and aspect ratio. Stored separately from the pixel count, which is why some files display stretched.
• Metadata. Title, creation date, sometimes camera details — mostly stripped by platforms on upload.
• Subtitle tracks. Possible in most containers, and platforms generally do not use them, which is why captions download separately — as text files.
• Chapters. Supported by some containers and rarely present in downloads.

Almost none of this survives a platform's upload processing, which is why a downloaded file tells you so little about where the video came from — what gets stripped.

How the parts fit together in practice

Six symptoms, three parts. Identifying which part is involved narrows the response immediately: an index problem is sometimes repairable, a codec problem needs a different player or a different download, and truncation needs the file fetching again.

What none of them needs is a conversion tool, which is the reflex people reach for and the one that helps least often.

Reading a file yourself

1. Windows: right-click, Properties, Details. Shows resolution, frame rate, duration and codec names.
2. macOS: open in QuickTime, Show Movie Inspector.
3. Any system: VLC's media information panel, which is the most detailed of the three.
4. Command line: ffprobe video.mp4 lists every track, codec and property.

Worth doing once on a file that is behaving strangely. Thirty seconds of looking usually replaces an hour of trying things, because the codec name explains the black screen and the duration explains the early stop.

The two numbers worth noting while you are there are the codec name and the duration. A codec you do not recognise explains a playback failure; a duration shorter than the source explains an early stop, and between them they cover most of the table above.

It is also the fastest way to settle an argument about whether a tool degraded something. Resolution and codec are facts about the file rather than impressions, and comparing them against what the platform offers takes less time than describing the problem to anyone.

Myths this structure debunks

Three widely held beliefs that dissolve once the three parts are clear in your head.

‘Converting to MP4 will fix it.’ Changing the container changes the box. If the problem is a codec the player cannot decode, the same codec inside a different box is still undecodable.

‘The file is corrupted.’ Usually it is not. Incomplete, or in an unsupported codec, or missing an index — three different states with three different responses, and only one of them involves damage.

‘A repair tool will recover it.’ It can rebuild an index. It cannot invent data that was never received, and any tool claiming to restore a truncated video is describing something that does not exist.

The common thread is that ‘broken video’ is not one condition. Knowing whether the container, a track or the index is the problem turns an unfixable mystery into a specific answer, usually a boring one.

A fourth belief is worth adding because it costs money rather than time. Paid repair software exists in volume, and what it does is rebuild indexes — a real capability, and one that is useless against the failure people usually have. Establishing whether the file is incomplete takes thirty seconds and decides whether any of it can help.

The test is the one from earlier: play the end. If the video simply stops before it should, the data is not there, and nothing sold anywhere will bring it back.