Can You Transcribe an Instagram Live?

Two different situations: your own Live, or someone else's

Which method you have depends on which side of the camera you were on.

• If you went live: the account that broadcast gets options once the stream ends — keeping the video on the device, or sharing the replay as a Reel. Either way you end up with a file you control, which is all a transcriber needs.
• If you watched someone else's Live: there is no download button. What you can get is the replay they choose to publish afterwards, usually posted as a Reel, or a screen recording made while the broadcast was still running.
• Screen recording is the universal fallback. It is the one method that exists for every Live regardless of what the broadcaster decides to do next, and the audio track is the part that matters for transcription.

The part nobody warns you about: a Live is too long to upload in one piece

A Reel runs for seconds. A Live runs for minutes to hours. That difference is the single biggest reason Live transcripts fail, and it has nothing to do with Instagram — it happens after you already hold the file. These are our own runs, same speech content, four lengths, ten uploads, all through the endpoint this site uses:

• 5 minutes, 879 KB: 3 of 3 runs succeeded — 39.3 s, 48.4 s, 41.3 s end to end.
• 10 minutes, 1.8 MB: 1 of 4 runs succeeded (87.1 s). The other three returned HTTP 500 after 91 s, 119 s and 131 s.
• 15 minutes, 2.7 MB: 0 of 2 — both failed after about 110 s.
• 30 minutes, 5.3 MB: 0 of 1 — rejected in 2.7 s.

Timings include the upload, measured on 2026-09-25. Note that the 30-minute file was refused almost instantly while the 15-minute file ran for nearly two minutes before failing, so those are two different walls rather than one. The practical guidance is the same either way: do not upload a whole Live.

Why no tool can fetch the Live from a link

Every tool that promises one-click transcription from a Live URL has to get the media first, and that is the part that does not work. These are the four paths a server-side tool would try, and what Instagram returns for each:

• Browser-side fetch: rejected by cross-origin-resource-policy: same-origin before the request reaches the media.
• yt-dlp: returns an empty media response without authenticated session cookies.
• oEmbed endpoint: requires an OAuth 2.0 access token; without one there is no video URL in the response.
• GraphQL endpoint: HTTP 400, invalid request, for unauthenticated calls.

How to cut a long recording into chunks that actually run

Five minutes is the length that did not fail once in our runs. Any tool that exports five-minute pieces will do the job; on a desktop, ffmpeg does it in one line. We ran this locally to confirm the syntax:

• ffmpeg -i live.mp4 -f segment -segment_time 300 -c copy -reset_timestamps 1 chunk_%03d.mp4
• Keep the pieces in one folder and name them in order — you will transcribe them one at a time and reassemble the text yourself.
• Any of these containers is accepted: MP3, WAV, M4A, AAC, OGG, FLAC, MP4, MOV, WEBM. Anything else comes back HTTP 400 with that list in the error message.
• If a chunk still fails, cut it shorter. Shorten before you assume something else is wrong.

Live audio is the hardest input, and here is how failure looks

A phone recording of a Live captures the room, not a studio: echo, a music bed if the broadcaster had one playing, notification sounds, and two people talking over each other. We do not publish an accuracy number because we have no test set to stand behind one. What we can tell you is exactly what our endpoint returns when a chunk goes wrong:

• HTTP 400, unsupported format: the container is not one of the nine. The message names them.
• HTTP 400, non-empty file required: the file is zero bytes. This is what a screen recorder that never actually saved produces.
• HTTP 422, the model returned no transcript: the chunk has no audible speech in it. Silence, or a music-only stretch, lands here.
• HTTP 500: the run failed. In our tests this is the code long files hit.

One more thing about screen recordings: viewer comments and questions are text drawn on the screen, not sound. They will not appear in the transcript, because they were never in the audio track.

What happens to each chunk after you upload it

The file is POSTed to /api/transcribe over an encrypted connection, decoded and processed in memory by @cf/openai/whisper running on Cloudflare Workers AI, and discarded as soon as the response goes back to you. There is no account, no file library, and nothing is retained for training. The flip side is that we cannot re-run a chunk on your behalf or hand you a copy later — keep your own recording and your own text.