M3U8 / HLS

Open an .m3u8 in a text editor and you will find a few dozen lines: a header, some parameters, and a long list of addresses. That is the whole file. The video lives elsewhere, cut into segments of a few seconds each, sitting on a server as thousands of separate small files.

The player reads the list, fetches the first few segments, starts playing, and keeps fetching ahead of itself. When your connection slows, it switches to a lower-quality variant listed in the master playlist — a different set of segments, at a different bitrate — and does it between segments so you see a quality change rather than a pause. This adaptive switching is the reason the format won.

It also explains something that surprises people: there is no single file to have. The film you are watching does not exist as an object anywhere. It is a thousand pieces and a set of instructions.

Because that is what you saved. The browser asked for the address you gave it, received four kilobytes of playlist, and wrote them to disk. Everything you actually watched arrived through a thousand separate requests that the save operation knew nothing about.

Downloading HLS properly means doing what the player does, but to disk: read the playlist, choose a quality variant, fetch every segment in order, and then stitch them into one container. That final step is called remuxing — the video and audio are not re-encoded, only rewritten into a single MP4 with a proper index so the result can be scrubbed.

Two things make this harder than it sounds. Segment addresses are often relative and change per session, so a list fetched an hour ago may be dead. And streams are frequently encrypted with a key referenced in the playlist, delivered separately, and rotated during playback.