IGTranscript

How many Instagram videos can I transcribe for free?

On this page there is no number, and that is the answer rather than a dodge: nothing here counts, so nothing here runs out. Free is not a quota on this page — it is the absence of every component a quota needs. The script that does the work is 22,349 bytes and contains zero localStorage, zero sessionStorage, zero indexedDB and zero document.cookie, so it cannot tell a first run from a five-hundredth. What limits a run is the machine you are sitting at, not a plan: the speech model is a one-time download of 79.7 MB (Better accuracy) or 43.6 MB (Smaller download), and each run holds the decoded audio in the tab at 26.88 MB per minute of 48 kHz stereo — 80.64 MB for a three-minute Reel. The three tools that rank above this page answer the same question with a daily number, with a credit balance, and with no free-allowance statement at all.

InVideo WW

What the three best-known tools say their free allowance is

The three pages that outrank this one for instagram transcript were fetched again on 13 October 2026, and each one answers the free-tier question in a different currency. One states a number of transcripts per day. One states a balance of credits and never converts it. One does not state a free allowance at all. Reading them side by side is the fastest way to see why "how many videos" is not one question with three answers but three different questions wearing the same words.

What does getthescript.app say the free tier is?

Its pricing page states the allowance in a single line: "Free: 3 short transcripts/day". The same page defines what a transcript costs against that allowance — "1 credit covers one video under 5 minutes" — and adds two rules that matter more than the headline: "Longer videos use +1 credit per 5-minute block" and "Failed transcript tasks cost no credit". So this is the only one of the three whose free number can be converted into videos, and only for clips under five minutes: one 20-minute Reel would consume four of the three credits that day, which is more than the daily allowance holds. The same 20-minute file costs this page no credits at all and instead holds 537.6 MB in the tab, which is the trade the two designs are making.

What does wayin.ai say the free tier is?

Its Instagram page states a balance rather than a rate: "Sign up to receive 100 free credits", repeated in the page's own FAQ as "You receive 100 credits when you sign up", with the closing note "no credit card required". Its newer Instagram video-to-text page, which appeared after 6 October 2026, says the same thing again — "You receive 100 credits when you sign up" — and still does not convert credits into videos anywhere on either page. That is the gap in this answer: a credit is a unit the tool defines, and without a stated price per video the balance does not tell you how many Reels it buys. On this page the comparable number is not a balance but a bill of materials: the export path is a 9,455-byte stylesheet and a 22,349-byte script, both served as fixed public files, and neither is counted against anything.

What does speakapp.com say the free tier is?

Its Instagram transcript page carries no free-allowance statement at all. What it says about cost is a claim about friction rather than a number: "Paste any Instagram Reel URL and get an accurate text transcript in seconds. No download or signup required." That is the whole of it — no transcript count, no credit balance, no reset period, and no currency symbol anywhere on the page. Whether that means unlimited or means an allowance it has simply not published is not something the page answers, and this page will not guess on its behalf. What can be said is what happens on the tenth Reel of a day here: it costs the same 26.88 MB as the first one, because there is no tenth-run price and no first-run price.

Why "free" here is not a quota

A limit is not a sentence on a page; it is machinery. Something has to know how many you have done, which means something has to write that number somewhere that survives a page reload, and something has to be able to refuse the next one. That is a chain of four links, and a page that is missing any one of them can publish the word "free" but cannot actually enforce a limit.

This page is missing all four, and the missing parts are countable. Its script is 22,349 bytes and contains 1 fetch( and 1 import(, both aimed at fixed public files — the recognition library at a pinned version and the speech model. It contains 0 FormData, 0 XMLHttpRequest, 0 sendBeacon and 0 POST requests, so it cannot construct a request that carries anything of yours. It contains 0 storage writes of any kind, so it cannot remember a count between visits. The front page carries 2 <input> elements — a file picker and a timestamp checkbox — and 0 text, email, password, number or telephone fields, so there is no field into which an account or a card could be typed. And across all 18 deployable files and 243,942 bytes there are 0 <form> elements, which is the element a sign-up would have to be built from.

Four things a page needs before it can limit how many videos you transcribe, drawn as a chain: an account to attach a count to, somewhere to write that count so it survives a reload, a server able to refuse the next request, and a number that eventually runs out. Under each link is the state of that link on this page. All four are absent. What has to exist before a page can limit you Four links in a chain. This page has none of them. 1. An account something to attach a running total to 2. Somewhere to write storage that survives a page reload 3. A server able to refuse the next request 4. A number that runs out absent absent absent absent Remove any one link and the other three are decoration. A page with no account, no storage and no server cannot hold a count to spend.
The chain is the argument: a quota is not a policy sentence, it is four pieces of machinery, and this page has 0 of the 4. The counts behind the four "absent" marks are 0 <form> elements across all 18 files and 243,942 bytes, 0 occurrences of localStorage, sessionStorage, indexedDB and document.cookie in a 22,349-byte script, 0 request-constructing primitives beyond the 1 fetch( and 1 import( that load the library and the model, and 2 <input> elements on the front page, both of which are a file picker and a checkbox.

What free actually costs: the download, then the memory

Free here does not mean costless, it means unbilled. Two costs are real and both are paid by your device. The first is the speech model, which is downloaded once per model per browser and then cached: 79.7 MB for Better accuracy and 43.6 MB for Smaller download. Measured on a browser that had never seen this page, the first visit with the smaller model transferred 45.9 MB and a repeat visit of that same model transferred 0 bytes. That is the shape of the cost — a one-time download, then nothing — and it is also why the honest answer to "how many can I do" starts with "how many models have you downloaded", not "how many videos have you done".

The second cost is memory, and unlike the download it is paid again on every run. Web Audio hands back decoded audio as 32-bit floats at the file's own sample rate and channel count, so one second of 48 kHz stereo is 48,000 × 2 × 4 = 384,000 bytes, and one second of the 16 kHz mono copy the model needs is 64,000 bytes. Both buffers exist at once while the resample runs, which is why a clip costs its source size plus a sixth of it rather than just its source size: one minute is 26.88 MB, three minutes is 80.64 MB, and ninety seconds is 40.32 MB. Inside a run the audio is also cut into 30-second slices of 480,000 samples each — 1.92 MB at 16 kHz mono — which is a property of the model call rather than a limit on the file.

The ceiling this produces was measured rather than assumed. On a two-core, four-gigabyte machine a 10-minute file decoded in 10.5 seconds and transcribed, and a 20-minute file was refused, because 20 minutes of 48 kHz stereo plus its resampled copy is 537.6 MB held at once. Length alone is not what decides it: a 30-minute file decoded cleanly when its audio was 16 kHz mono, a sixth of the size. Two smaller files also went through without trouble — a 6.3 MB, 43-second MP4 and a 1.3 MB, 161-second MP3.

So the answer to "how many" is not a total you spend down; it is the size of the largest single file your tab can hold, applied once per run. Running two short clips is cheaper than running one long one, and running a hundred short clips is exactly as cheap as running one, because the peak is set by the biggest file in the batch and not by the batch.

Thirty short Reels in one sitting: the arithmetic

Take thirty Reels of thirty seconds each, which is fifteen minutes of audio spread across thirty files. The model is downloaded once: 79.7 MB for Better accuracy, 0 bytes on every run after that. Each run then peaks at 13.44 MB, because a 30-second clip at 48 kHz stereo is 11.52 MB of source plus 1.92 MB of resampled mono, and the page accepts one file at a time so those peaks do not add up. The decoded audio is not carried between runs either — the result the page keeps holds the text, the segments, the duration and the timings, and not the samples — so the thirtieth run holds the same 13.44 MB the first one did. On memory alone, thirty short Reels cost less than one 20-minute Reel, which would have asked for 537.6 MB and been refused.

Thirty short Reels against one long one. Along the top, thirty squares stand for thirty thirty-second Reels, which is nine hundred seconds of audio spread across thirty files. The long bar is a single twenty-minute Reel at 48 kHz stereo, which asks for 537.6 MB at once and was refused on a four-gigabyte machine. The tiny bar underneath is what each thirty-second Reel costs on its own, 13.44 MB, and because the page accepts one file at a time those peaks never add up. Thirty short Reels cost less than one long one The page takes one file at a time, so the peaks never add up. thirty 30-second Reels · 900 seconds of audio · one at a time one 20-minute Reel at 48 kHz stereo 537.6 MB 537.6 MB held at once — refused on a four-gigabyte machine each 30-second Reel, held on its own 13.44 MB 11.52 MB of source plus 1.92 MB resampled — and thirty of them never add up The decoded audio is not carried between runs, so the thirtieth run holds what the first one did. On memory alone, thirty short Reels cost less than one 20-minute Reel — a factor of 40.
Thirty Reels of thirty seconds each is fifteen minutes of audio spread across thirty files. Each run peaks at 13.44 MB, because a 30-second clip at 48 kHz stereo is 11.52 MB of source plus 1.92 MB of resampled mono, and the page accepts one file at a time so those peaks do not add up — the decoded audio is not carried between runs either, so the thirtieth run holds the same 13.44 MB the first one did. One 20-minute Reel at 48 kHz stereo asks for 537.6 MB at once and was refused on a four-gigabyte machine, which is 40 times as much. That arithmetic is derived from the buffer sizes and the structure of the script rather than from a measured thirty-file session; what would actually stop you is time, since thirty Reels of thirty seconds is 900 seconds of audio and a 161-second clip took 112 seconds here.

That arithmetic is derived from the buffer sizes and from the structure of the script rather than from a measured thirty-file session, and it is worth saying which is which. What would actually stop you is time, not a limit: transcription runs at roughly half to two-thirds of the audio's length on a modest machine, and the spread is wide — measured here, 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, and a 161-second clip took 112 seconds. Thirty Reels of thirty seconds is 900 seconds of audio, and the honest range that the measured spread implies is several minutes of waiting to a quarter of an hour, with nothing counting in the background while it runs.

How to check any free-tier claim yourself

  1. Look for the counter before you look for the number. A published allowance of any size needs storage and an identity to attach it to, so count the login forms and the storage calls on the page. This page comes out at 0 <form> elements and 0 storage writes in 22,349 bytes of script, which is why it can say "free" without saying "how many".
  2. Read what one credit buys, not how many credits you get. A balance in an undefined unit is not an answer to a question about videos. getthescript.app is the only one of the three that publishes the conversion — "1 credit covers one video under 5 minutes" — and once you have it, their "3 short transcripts/day" becomes a real number while wayin.ai's 100 credits still does not.
  3. Watch the second run's network traffic. If a free tier is really a one-time download rather than a per-run cost, the second run should transfer nothing. Measured here, a repeat visit with the same model transferred 0 bytes against 45.9 MB on the first visit, and the whole page is 243,942 bytes of which 22,349 is the script and 9,455 the stylesheet.
  4. Check whether a card can even be entered. The strongest evidence that a charge cannot appear is that no field exists to type a card into. This page has 2 <input> elements, a file picker and a checkbox, and 0 text, email, password, number or telephone inputs anywhere on it.

Will it suddenly start charging me?

Not on this page, and the reason is structural rather than a promise. There is no account to upgrade, no card field to store, 0 <form> elements to submit and 0 storage writes to read a previous purchase back from. There is also no backend: the only two requests the page makes are GETs for fixed public files, so there is no server of ours that could hold a balance or bill one. The tool either runs entirely on your device or it does not run at all.

What can genuinely go wrong is different, and it is worth being exact about it. Those two GETs go to third parties — the recognition library from cdn.jsdelivr.net, pinned at @huggingface/transformers@3.7.5 so that the library cannot change under the page without this line changing too, and the speech model from huggingface.co. If either host is unreachable, the page reports that it could not load and does nothing else; it does not fall back to a server, because there is none to fall back to. The library is about 4 MB and is cached like any other web file. The failure mode of a design with no billing is an outage, not an invoice.

Last checked: 13 October 2026. The counts of this site's own files were taken on this machine on that date rather than remembered: the script is 22,349 bytes, the stylesheet is 9,455 bytes, and all 18 deployable files come to 243,942 bytes, with 0 <form> elements among them and 0 occurrences of localStorage, sessionStorage, indexedDB or document.cookie in the script. The two model sizes, 79.7 MB and 43.6 MB, and the measured 45.9 MB first visit and 0-byte repeat visit, are from this tool's own recorded measurements. The Instagram-side quotations are the three tools' own sentences, fetched on 13 October 2026, and are reproduced as sentences rather than paraphrased. The audio arithmetic is derived rather than measured — Web Audio fixes a decoded sample at 32 bits, so 48,000 × 2 × 4 = 384,000 bytes per second of 48 kHz stereo and 16,000 × 4 = 64,000 bytes per second of 16 kHz mono, and both buffers exist while the resample runs, giving 13.44 MB for a 30-second clip, 26.88 MB for a minute, 80.64 MB for three and 537.6 MB for twenty. The refusals and the timings were measured in an ordinary browser tab on a two-core, four-gigabyte machine. Figures that could not be measured or derived are absent rather than estimated, and no allowance, price or traffic number for any other tool is inferred here.

If the thing you are actually deciding is whether to run a stack of files rather than one, bulk transcribing Instagram Reels works through what changes when there is more than one. If the concern behind the question is what you have to hand over to use a tool at all, transcribing with no login compares what each of the three best-known tools actually hands an anonymous visitor, and whether transcribing uploads the video covers where the work runs. If a run has already failed, the five failure modes is the page to start from, and if the file is too long to finish, the ceiling is explained on the Reel length limit page.