How do you transcribe an Instagram carousel?

You cannot transcribe a carousel as one object, because a carousel is not one file — Instagram lets a single post carry up to 20 photos and videos, and every transcription tool, this one included, reads one file's audio track at a time. The part that is transcribable is the video slides, and only the ones with speech in them; a photo slide has no audio track in it at all, so there is nothing in it for any tool to read. The part that surprises people is the memory arithmetic. Twenty slides of 60 seconds each hold 460.8 MB of decoded audio between them — exactly the same as the single 20-minute file this tool refused on a four-gigabyte machine — and the carousel goes through anyway, because it arrives as 20 separate decodes of 23.04 MB rather than one decode of 460.8 MB.

• Only the video slides carry anything you can transcribe, and only the ones with speech in them. The tool hands the whole file to the browser and asks for its decoded audio, which comes back as 32-bit samples at the file's own rate and channel count — 384,000 bytes for every second of 48 kHz stereo — and resamples that down to 16,000 samples per second of mono at 64,000 bytes a second. The frames are gone by then, so a slide whose words are burned into the picture contributes nothing at all. This tool hears; it does not watch.
• A photo-only carousel is not a hard case, it is an empty case, and it never downloads the model. The decode step runs before the model is fetched, so a slide with no audio track fails in about a second instead of after the 79.7 MB download, and the message names all three causes the browser cannot tell apart. If every slide in a carousel is a photograph, you can stop looking for the setting that would make it work — there is no audio to find.
• One file at a time, and the picker cannot be talked into more. The file input carries a fixed 12-entry accept list and no multiple attribute, and the change handler reads the first file in the list and ignores the rest — so dropping a whole carousel onto the box transcribes exactly one of its slides and says nothing about the other 19. A whole carousel dropped at once therefore costs exactly one slide's decode, 23.04 MB at 60 seconds, no matter how many slides the post contains. The drop zone behaves the same way, which is the one place this page would have expected the two paths to differ.
• The naming trap is the part that costs you the most, and it is entirely on this page. Every download is written to a fixed name — transcript.txt, transcript.srt, transcript.vtt — and the source file's own name is recorded on the result object and read by 0 of the export paths, so a 20-slide carousel gives you 20 files called transcript.txt and nothing inside any of them that says which slide it came from. Rename each one the moment it lands.
• A carousel is structurally safer than one long clip, which is the opposite of what people assume. Because each slide is its own run, the ceiling that decided the 20-minute file — the decoded audio held in memory — applies to the largest single slide rather than to the post. A full 60-second carousel slide decodes to 23.04 MB of 48 kHz stereo, which is one twentieth of the 460,800,000-byte decode that was refused, and 3.84 MB once it has been resampled to 16 kHz mono.

Why twenty 60-second slides go through where one twenty-minute file does not

The arithmetic is not this tool's; it comes out of the Web Audio specification and it is the same arithmetic everywhere. A decoded sample is 32 bits, so one second of 48,000 Hz two-channel audio is 48,000 × 2 × 4 = 384,000 bytes, and the browser hands that buffer back before any resampling happens. Twenty minutes of it is 1,200 × 384,000 = 460,800,000 bytes, or 460.8 MB, and that single decode is what the four-gigabyte machine refused. Sixty seconds of the same audio is 23,040,000 bytes, or 23.04 MB. Twenty 60-second slides are twenty of those, and 20 × 23,040,000 = 460,800,000 bytes again — exactly the same figure, split twenty ways. That equality is the whole reason a carousel is easier than a long clip rather than harder.

Nothing on this page queues slides for you, and that is the honest limit of the claim: the reason the carousel goes through is that each slide is decoded inside its own run, and the previous slide's buffer is not referenced by anything once that run returns. The result object keeps the text, the segments, the duration, the elapsed time, the model, the language and the file name — it does not keep the audio. So the peak here is one slide's decode rather than the carousel's sum, and 3.84 MB per slide is what actually sits in memory while the model runs. Twenty slides that are all photographs never get that far at all.

The other number that moves is time, and no arithmetic shortens it. A 60-second slide is two of the model's 30-second windows, so a full 20-slide carousel is 40 windows, and transcription runs at roughly half to two-thirds of the length of the audio on a modest machine. On 1,200 seconds of video that is somewhere between 10 minutes and 13 minutes of waiting at the machine's own pace, and longer on a phone. If the carousel is four video slides rather than twenty, it is the arithmetic that shrinks, not the patience required.

What the three best-known tools say about carousels

The three pages that outrank this one for instagram transcript were fetched again on 15 October 2026. The word carousel appears 0 times across all three of them, and so do slide and multi-image. The nearest anyone comes is wayin.ai, which mentions carousels 14 times — 7 on one page and 7 on the other — and every single one of those is either a reach statistic or a rule about Carousel Ads. Not one is about transcription. What any of the three does when you hand it a link to a post that contains several videos was not tested here, and this page is not going to guess it.

What does getthescript.app say about carousels?

Nothing, and the gap is a specific one rather than a vague one. Its Instagram page offers three routes in — a single video, a batch mode and an upload — and every heading on it is about one Reel or one URL. There is no fourth route for a post containing more than one video, and no mention of a carousel, a slide or an album anywhere in it. That batch mode is the interesting near-miss: it is built for twenty URLs, not for one URL that turns out to be twenty files, and the page does not say which slide a carousel link resolves to — the first, the last, or all of them. This page cannot measure that from the outside, so it is recorded here as unknown rather than assumed.

What does speakapp.com say about carousels?

Nothing either, and its stated ceiling makes a carousel interesting rather than easy. It says it supports public Instagram videos and that longer videos are trimmed to 2 minutes, and elsewhere that it supports all public Reels up to 2 minutes long — so a carousel's 60-second video slide fits under that ceiling comfortably, which is a point in its favour. But it takes a URL, and a carousel's URL is the URL of a post, not of a video. What it does with a post URL that happens to carry three videos is not observable without an account, so it is not claimed here.

What does wayin.ai say about carousels?

wayin.ai is the only one with carousel numbers at all, and they are all about reach and ads. Its Reels-creation page states that Reels see 2.25x more reach than single photo posts and 36% more reach than carousels, and that carousels see 12% more engagement than Reels and 114% more than single image posts. Its length page states “Carousel Video Limits: 60 Seconds Per Video” and that you can include between 2 and 10 cards per carousel — the only slide count any of the three gives, and a Carousel Ads figure rather than an organic-post one. Its own Instagram transcript page does not mention carousels at all.

That 10-card figure is where checking pays for itself. Metricool reported on 9 August 2024 that Instagram raised the carousel limit from 10 to 20, and a second Metricool guide dated January 2026 states the limit as up to 20 photos, videos or a mix of both. MacRumors' Instagram guide records the same change. Instagram's own help centre could not be reached from this machine when it was tried on 15 October 2026 — two separate help-center URLs returned “This Page Isn't Available” — so the 20 figure rests on those three secondary sources rather than on Instagram itself, and it is safest written down as “up to 20” rather than as an exact number.

The order to do them in

1. Count the video slides before you start. Swipe through and note which ones have a play triangle or a sound indicator. That is your run count and the only number that matters for the work: a carousel with three videos in it is three runs, and the thirteen photographs around them cost you nothing except the time it took to count. Each 60-second slide is a self-contained unit of cost at 23.04 MB of decoded audio and two of the model's 30-second windows, so there is no wrong order to do them in.
2. Transcribe them in the order they appear, one run each. Nothing carries over between runs — the language picker keeps your setting, and the previous slide's audio is not kept. The one thing that does carry over is the cached model, which matters: only the first video slide of the carousel pays the 79.7 MB download, and every slide after it reads the same model from the browser's cache, so slide 19 costs slide 1's patience and nothing else.
3. Rename the file the moment it lands, before you pick the next slide. The download is written to transcript.txt regardless of what the slide is called on Instagram, and the source name is stored on the result object where no export path reads it, so the only thing that keeps a twenty-slide carousel straight is you naming each piece as it arrives.
4. Do not join the parts by concatenating the subtitle files. Every slide's cues start at 00:00:00,000, because the stamp is built from the timestamps the model returns for that one file and every file starts at zero. Twenty SubRip files glued together give you a file whose cue numbers run from 1 to about 2 twenty times over, and whose timestamps all claim the beginning of the timeline — which plays as twenty subtitles stacked on the same instant. Stitch the text, not the timestamps, or renumber by hand.

What a carousel cannot give you

Three things are outside what any transcript of a carousel contains. It cannot say who spoke, because the call that produces it passes four options — language, task, return_timestamps and chunk_length_s — and 0 of the four has anything to do with separating voices, so a carousel of two people talking over dinner comes back as one undifferentiated column. It cannot read text that was burned into a photograph, because a photograph is not audio and never reaches the decoder at all. And it cannot hand you the post as one object: a carousel stays a pile of separate files, on the Instagram side and on this one, and the moment you want them combined you are doing the combining.

What it does give you is a set of things you can act on. A carousel whose slide 4 explains a method is as searchable as anything else on this page once slide 4 has been transcribed, and a slide with no speech in it produces a panel that says so rather than a file full of guesswork. The honest summary is that the question is not whether a carousel can be transcribed — its video slides can, at the same quality as any other clip of the same length — but that the unit of work is the slide, and nothing anywhere in this pipeline is going to change that for you.

Last checked: 15 October 2026. The memory arithmetic on this page is derived from the Web Audio specification rather than measured: a decoded sample is 32 bits, so 48,000 × 2 × 4 = 384,000 bytes per second of 48 kHz stereo, 60 seconds is 23,040,000 bytes (23.04 MB) and 1,200 seconds is 460,800,000 bytes (460.8 MB); the 16 kHz mono figure is 16,000 × 1 × 4 = 64,000 bytes per second, so 60 seconds is 3,840,000 bytes (3.84 MB). Every one of those products was printed and re-checked before it was written down, and the equality 20 × 23,040,000 = 460,800,000 was confirmed rather than estimated. The 79.7 MB model size, the 43.6 MB alternative, the 45.9 MB first visit and the two interviews required to confirm the refused 20-minute file are this tool's own recorded measurements, taken in an ordinary browser tab on a two-core, four-gigabyte machine. The 22,349-byte script and the 11,749-byte stylesheet were measured on this machine on that date, as were the 12 entries in the file input's accept list, the single files[0] read in both the change handler and the drop handler, the one text/plain;charset=utf-8 string, and the 30,000-millisecond delay before the blob URL is revoked. The 60-second-per-slide figure is quoted from wayin.ai's own length page, and the 20-slide ceiling is reported by Metricool (August 2024 and January 2026) and MacRumors — Instagram's own help centre did not load when it was tried on 15 October 2026, so no Instagram URL is cited for it. The carousel counts of 0 and 14 were counted over the 39 files of competitor copy in this site's own benchmark directory, not over the internet. What any of the three tools does with a carousel link was not tested and is not claimed.

If the question is actually about a batch of these rather than one post, bulk transcribing Instagram Reels covers what changes when there is more than one file and what the model download costs on the first of them. If the ceiling itself is what you are up against, the Reel length limit page has the decoded-audio arithmetic behind it. If the slides keep coming back empty rather than short, the five failure modes is the page to start from, and if the words are right but you want to know what to build out of them, what to do with an Instagram transcript has the four outputs and their byte costs. If what you care about is whether the video leaves your machine at all, whether transcribing uploads the video answers that one.