Instagram Reel length limit

Instagram caps how long a Reel may be; that cap is not the one that stops a transcript. The limit you actually hit when you want the words out is memory. This tool holds the whole decoded audio track in the tab before it transcribes a word, and decoded audio costs 384,000 bytes for every second of 48 kHz stereo — so a three-minute Reel that arrives at 48 kHz stereo occupies 69.12 MB, while the same three minutes arriving as 16 kHz mono occupies 11.52 MB. Both are ordinary Reels, both transcribe identically, and only one of them is near a problem. There is no length limit in the code at all: the only number in the source that has anything to do with duration is chunk_length_s: 30, which cuts the audio into 30-second pieces and gives a three-minute Reel 6 chunks to work through.

Instagram's own cap is harder to pin down than the guides suggest. Two official help pages were read on 7 October 2026 — the Reels page and the Reel size and aspect ratios page — and neither states a duration at all. Between them they give an aspect ratio of 1.91:1 to 9:16, a minimum frame rate of 30 FPS and a minimum resolution of 720 pixels, and nothing whatsoever about how long a Reel may run. The figure most often repeated elsewhere, three minutes, and the longer figure that began circulating during 2026, could not be confirmed on an Instagram help page either, so this page does not present either of them as official. What it can do instead is tell you exactly what happens when a given length is fed to this tool — measured, and worked out from the page's own source, in the sections below.

Last checked: 7 October 2026. The memory figures on this page were derived from the page's own source code and from the Web Audio specification, and each one was printed and checked in Python before it was written down; the timings and the pass-or-refuse results were measured on this tool, on a two-core, four-gigabyte machine, in a normal browser tab. Nothing here is estimated, and where a number could not be derived it is left out rather than guessed.

The limit that actually stops a transcription

Nothing on this page counts seconds, and the browser does not either. Before any speech recognition runs, the file's audio track is decoded into raw samples and held uncompressed, and that buffer — not the duration, and not the size of the file on disk — is what has to fit in memory. Those samples are 32-bit floating point at the file's own sample rate and channel count, so the arithmetic is fixed: bytes per second is sample rate × channels × 4. Written out for the formats a Reel is likely to arrive in, that is 384,000 bytes a second (23.04 MB a minute) for 48 kHz stereo, 352,800 bytes a second (21.17 MB a minute) for 44.1 kHz stereo, 128,000 bytes a second (7.68 MB a minute) for 16 kHz stereo, 64,000 bytes a second (3.84 MB a minute) for 16 kHz mono, and 32,000 bytes a second (1.92 MB a minute) for 8 kHz mono.

Which bar you get is decided by the sample rate and channel count of the file you choose, and nothing visible in a file browser tells you which one you have. A Reel recorded on a phone sits near the top; a Reel whose audio has already passed through a converter, or a voice memo, sits near the bottom.

The spread is twelve to one for the same three minutes, and it is the single most useful thing to know about this topic. A Reel recorded on a phone is usually 44.1 or 48 kHz stereo, which is the expensive end of that list. A Reel that has already been through an editor, or audio that started life as a voice memo, is often 16 kHz mono, which is the cheap end. The two files look identical in a folder, the duration is the same, and the transcript is the same — only the memory is different.

Then there is a second buffer, and it is easy to miss. The recognition model wants 16 kHz mono, so the page resamples the decoded audio down to exactly that: 64,000 bytes a second, or 11.52 MB for a three-minute Reel, no matter what the source was. The resample runs while the original buffer is still alive, so for a moment the tab is holding both. For a three-minute Reel that arrived at 48 kHz stereo that is 69.12 MB plus 11.52 MB — a peak of 80.64 MB, not 69.12 MB. The peak, not the average, is where a tab dies.

What a three-minute Reel actually costs

Three minutes is worth costing out precisely, because it is the length most Reels are near. All of the following comes from the page's own source and from arithmetic that follows from it:

The model works in 30-second pieces. The source sets chunk_length_s: 30, so a three-minute Reel is 6 chunks and nothing about the result depends on where those boundaries fall — the exported timestamps are measured from the start of the whole file, not from the start of a chunk.
The decoded buffer. 69.12 MB if the Reel arrives at 48 kHz stereo; 11.52 MB if it arrives at 16 kHz mono. Nothing in between is unusual.
The resample buffer. 11.52 MB, always, because the target is fixed at 16 kHz mono (RATE = 16000 in the source) regardless of the input.
The peak the tab has to survive. 80.64 MB from a 48 kHz stereo Reel, or 23.04 MB from a 16 kHz mono one — the sum of the two buffers above, held at the same moment.
The download, once. 79.7 MB for the default model on first use, or 43.6 MB for the smaller one. After that the browser serves it from cache, so the second Reel costs no bandwidth at all. The two model files behind the default are encoder_model_quantized.onnx at 23.20 MB and decoder_model_merged_quantized.onnx at 53.69 MB.
The wait. Transcription runs at roughly the speed of playback on a modest machine. Measured here: a 161-second clip took 112 seconds, so a three-minute Reel — 19 seconds longer than that clip — is in the same territory. A 43-second clip took anywhere from 18 to 42 seconds across a dozen runs, which is the machine being busy rather than the model being unpredictable.

Put together, a three-minute Reel is not a hard case. Its peak is 80.64 MB in the worst common format, the model has already been downloaded by the time you are on your second one, and the wait is around two minutes. If three minutes were the only length anyone transcribed, this page would not need a limits section at all.

Where the length limit really bites

It bites when the clip is not a Reel. The measured evidence on this machine is a 30-minute file that decoded and transcribed cleanly, and a 20-minute file that was refused. The 30-minute one went through because its audio was 16 kHz mono: 115.2 MB decoded, 230.4 MB at peak. That is the largest decode this tool is known to have completed, and it is a useful line to hold on to, because it is a real measurement rather than a guess. The 20-minute file that was refused cannot be converted into a number at all — its sample rate and channel count were not recorded when it was tried, so either reading of it (460.8 MB if it was 48 kHz stereo, 76.8 MB if it was already 16 kHz mono) fits the evidence, and inventing one of them would be a lie.

Against that line, the longer formats stop looking comfortable. A feed post may run far longer than a Reel, and the arithmetic scales straight up with duration:

A three-minute Reel sits far to the left of this chart — 69.12 MB in the worst common format, well under the line. Everything that goes wrong with length goes wrong out here, in the formats that are not Reels.

The same 60 minutes, arriving as 16 kHz mono, decodes to 230.4 MB with a peak of 460.8 MB — still double the largest decode known to have completed, and still the optimistic case. Arriving as 48 kHz stereo, which is what a screen recording of a desktop or a phone capture normally gives you, the same 60 minutes decodes to 1,382.4 MB with a peak of 1,612.8 MB. That is roughly twelve times the largest decode measured passing on this machine, and it is not a figure this page will pretend to be comfortable with. The honest statement is that a 60-minute file in 48 kHz stereo is beyond what has been shown to work here, and that a 60-minute file in 16 kHz mono is unproven rather than safe.

This is the part the other tools' length answers skip. Saying "no matter the length" is not a claim about a limit; it is a refusal to name one. The limit is real, it is measurable, and it has nothing to do with how many minutes the platform will let you post.

What do you do when a clip is over the limit?

You change the file's sample rate before you change anything else, because that is where the factor of six lives: the same three minutes decodes to 69.12 MB at 48 kHz stereo and 11.52 MB at 16 kHz mono, and the peak falls from 80.64 MB to 23.04 MB with it. Shortening the clip and splitting it into pieces work too, and any of the three keeps the peak under the 80.64 MB measured above. The four moves, in the order that costs least effort:

Change the format before you change the length. A three-minute Reel at 48 kHz stereo costs 69.12 MB decoded and 80.64 MB at peak; the identical three minutes at 16 kHz mono costs 11.52 MB and 23.04 MB. That is a factor of six, for a transcript that comes out the same, because the model resamples to 16 kHz mono internally anyway. Any audio editor will export 16 kHz mono, and it is the single highest-value thing on this page.
Shorten the clip, if you can. Duration scales the buffer linearly: a 90-second clip at 48 kHz stereo decodes to 34.56 MB (peak 40.32 MB), and the same 90 seconds at 16 kHz mono decodes to 5.76 MB (peak 11.52 MB). If the words you need are in the first minute, do not decode the ninth.
Split it into pieces. The model already works in 30-second chunks, so cutting a long file into two- or three-minute parts changes nothing about the output and keeps every run under the 80.64 MB peak measured above. Export each piece as you go — this page keeps no history, so an unexported transcript is gone when you move on.
Do the long one alone. The tab competes with every other tab for the same memory, and the peak is the sum of two buffers rather than one. If the tab runs short it can be killed outright, and a killed tab cannot show you an error — you would simply see it stop.

The three answers this question gets elsewhere

For comparison, and quoted exactly as published on 6 October 2026:

speakapp.com names a figure, and it is a short one. Its Instagram transcript page says: "Supports public Instagram videos (longer videos trimmed to 2 min)", and in its own steps: "We support all public Reels up to 2 minutes long." (speakapp.com, /t/tools/instagram-transcript.) That is the most honest of the three, and it also means a three-minute Reel is not fully transcribed there. For scale, the two minutes it keeps is 46.08 MB of decoded audio at 48 kHz stereo (53.76 MB at peak), against the 69.12 MB those same three minutes cost here.
wayin.ai names no figure at all. Its Instagram tool page says: "Paste the public link from a Reel, a Story, or an hour-long Live replay—WayinVideo transcribes every spoken word with precise timestamps, no matter the length." (wayin.ai, /tools/video-transcript-generator/instagram/.) "No matter the length" is the claim; the number behind it is not given, and no machine has an unlimited buffer.
getthescript.app says nothing about length on its Instagram transcript page — not a limit, not a range, not a "no limit". The absence is the answer.

Three questions will get you past all of that, whichever tool you end up using. Does it name a number of minutes, and is that number a limit or a truncation? If it says the length does not matter, ask what happens to a 60-minute file in 48 kHz stereo — the answer will either be a number or a change of subject. And ask where the work happens, because a tool that fetches the video to a machine it controls has a server's memory to spend and can honestly accept longer files than one that runs in your tab.

Where this leaves you

For the question as people actually ask it — how long can a Reel be, and can I get the words out of it — the answer is that a Reel's own cap is not the binding constraint. A three-minute Reel is a small job here: 6 chunks, 69.12 MB decoded in the worst common format, 11.52 MB in the best, a peak of 80.64 MB, and a wait of about two minutes. The binding constraint shows up in the longer formats, where the decoded audio runs from 230.4 MB at 60 minutes in 16 kHz mono to 1,382.4 MB in 48 kHz stereo, and where the largest decode this tool has actually completed is 115.2 MB. If you are near that line, change the sample rate before you change anything else.

Everything above describes the same tool as the home page — the file picker, the language setting and the export buttons. Limits has the file-size and speed measurements, bulk transcribing Instagram Reels covers what changes when there is more than one file, and the privacy page covers what does and does not leave your device.