Twitch Stream DownloaderRecord a live Twitch stream to MP4 — as it happens.

A live capture starts at the moment you press Download, not at the start of the broadcast. Twitch's live playlist advertises only a rolling window of the most recent segments, so the recording opens with whatever is in that window and then moves forward in real time. Anything that aired earlier is not in the playlist and cannot be fetched. To get the earlier part you have to wait for the past broadcast and download that instead, which only works if the channel saves its VODs and only while Twitch still keeps them.

While recording, each segment is written out in the raw MPEG-TS form Twitch delivers it in, with no conversion stage. The MP4/TS format switch does not apply to a live capture: it always finishes as a .ts file, and an audio-only live capture as .aac. That is a container difference, not a quality difference, since the video and audio pass through untouched. VLC plays .ts as-is, and ffmpeg -i in.ts -c copy out.mp4 rewraps it into MP4 without re-encoding. Downloading the past broadcast afterwards is the route to a direct MP4, because the MP4 format option does apply there.

Between segments the recorder waits the interval Twitch advertises in the playlist (EXT-X-TARGETDURATION, or two seconds if the playlist does not state one), re-reads it, and appends any segment it has not already seen. Recording ends on one of four things: you press Stop, the playlist signals the broadcast is over (EXT-X-ENDLIST), eight consecutive polls bring back nothing new or fail outright, or the three-hour per-session cap is reached, which exists as an abuse limit. After that you can start a new recording for another three hours. Since segments are appended as they arrive and nothing is converted at the end, whatever was captured before it stopped is playable.

Where those bytes go while recording depends on the browser. Chrome and Edge can stream to disk: they ask for the save location before the first segment arrives, so within the three-hour cap the practical limit is free disk space. Firefox and Safari have no equivalent here, so the capture is held in memory until you stop it, which is a real constraint over a multi-hour session. Either way there is no ETA or percentage, because the end point is unknown; the progress bar simply pulses while the readout counts segments captured, total bytes and current speed.