Instagram Reel Transcript Generator: One Reel, One 147-Word Sentence
Five public Reels went through the same generator on the same afternoon with the same settings. One came back as a single 147-word sentence with no full stop in it; another as 39 sentences averaging 6.7 words. The punctuation is the model's, and it moved 22-fold between clips.
Main guide: Instagram transcript — paste a link, get the full text. The page you are reading covers one specific case in depth.
Short answer: five Reels, 997 words, 94 sentences averaging 10.6 words — and that average hides the spread. Sentence density ran 0.68 to 15.09 per 100 words, one clip carried no punctuation at all, and five sentences held 39% of the words. Before quoting, count full stops: under 5 per 100 words the text is unbroken, over 13 it is fragments.
Visual guide: One reel returns 0.68 sentences per 100 words
Three slides: The other returned 15.09. Same generator, same settings, one afternoon: sentences per 100 words, five Reels.



Watch the guide
Three slides, thirty seconds, English captions in the player. The video is silent by design; the captions carry the words.
Five Reels, one engine, one afternoon
All five were public and English, run once each on 27 October 2026 through the engine this site uses: base model, beam 5, timestamps off. Length is the decoded audio. The last column counts sentences longer than 25 words.
| Clip | Length | Sentences | Words per sentence | Longest sentence | Over 25 words |
|---|---|---|---|---|---|
| DewABZzkTMb | 135.32 s | 39 | 6.7 | 21 | 0 |
| DdynUVJhJRW | 92.46 s | 5 | 44.6 | 114 | 3 |
| De8C3dYFQbZ | 55.57 s | 24 | 6.6 | 33 | 1 |
| DeWO-G0ivEG | 51.69 s | 25 | 8.2 | 21 | 0 |
| DLUOQr6JLY4 | 44.03 s | 1 | 147.0 | 147 | 1 |
379.07 seconds of audio, 997 words, 94 sentences. Mean 10.6 words per sentence; the median for each clip in the table order was 6, 42, 4, 7 and 147.
The clip that came back with no punctuation
DLUOQr6JLY4 runs 44.03 seconds and returned 147 words as one sentence: zero commas, zero question marks, zero capitals, no full stop at the end. There is no break in it to move. Pasted into a document it is one line that has to be punctuated by hand, and the marks you add are yours, not the speaker's.
It is not an artefact of one bad clip. Three of the five ended with a full stop and two did not, and only two began with a capital. The edges tell you which kind of output you are holding.
The clips that came back in pieces
At the other end, 71 of the 94 sentences (75.5%) are ten words or shorter, 25 are three words or shorter, and 8 are a single word. DewABZzkTMb turned 263 words into 39 sentences with a median of 6 words. De8C3dYFQbZ split 159 words into 24 sentences, median 4.
Fragments read more easily than a 147-word run and reuse worse. A phrase that crosses a break will not be found by a search for the phrase, and a quote lifted at a break is a fragment because that is what it is.
Where the words actually sit
Five sentences out of 94 run past 25 words, and they hold 389 of the 997 words: 39% of the output. On DdynUVJhJRW three of its five sentences are over 25 words and the longest is 114. The text is mostly short units with a few long runs, and the long runs are where the argument is.
A written paragraph is a handful of sentences of similar length. This is a scatter: 8 one-word units and one 147-word unit from one afternoon.
The ten-second test before you quote
Count the full stops in the returned text, divide by the number of words, and multiply by 100. That is the sentence density, and it sorts the output into three cases you handle differently.
| Full stops per 100 words | What you are holding | What to do |
|---|---|---|
| Under 5 | One unbroken run, possibly with no punctuation at all | Break it yourself, and say the breaks are yours |
| 5 to 13 | Sentences close to the length a reader expects | Quote as it stands, after reading the longest one |
| Over 13 | Fragments, many of them three words or fewer | Join them before quoting, or quote the fragment as a fragment |
Four of the five clips landed outside the middle band: 0.68, 2.24, 14.83, 15.09. One landed inside it, at 12.20. Run one Reel and the odds are not on the middle band, which is why the count is worth ten seconds.
Why the breaks move, and what is not adjustable
Nothing in this pipeline chooses sentence boundaries. The model is asked for text only: return_timestamps: false at public/app.js line 147 requests one string instead of timed segments, and line 149 trims it. No setting on the page touches sentence length or punctuation.
So the breaks are whatever the model produced for that audio. Five clips cannot say why one came back unbroken and another in 39 pieces; this page claims no cause, only that the difference is real and measurable.
What to do with the text you get
Under 5, do not quote a sentence, because there is not one: quote a clause and mark the cut. Over 13, join the fragments back into the sentence the speaker spoke before the text goes into a caption. In the band, still read the longest sentence, because a 114-word sentence quoted whole is a paragraph nobody finishes.
One more check costs nothing: the last character. Three of the five texts ended with a full stop and two did not; a text that stops mid-thought is one where the tail may not have arrived.
Questions
- Does the generator add full stops?
- Sometimes. Three of the five ended with one and two did not, and one 44-second clip came back as 147 words with no comma and no full stop anywhere.
- How many sentences should a one-minute Reel produce?
- At 9.4 sentences per 100 words, a 600-word transcript lands near 56. The per-clip spread was 0.68 to 15.09, so plan for a range, not a number.
- Can I make it break sentences where I want?
- No. The call asks for one string with
return_timestamps: falseatpublic/app.jsline 147, and no sentence-length setting exists. Breaks are added after the text arrives, and then they are yours.
Affiliate link: Notta is a separate service that transcribes audio and video files you upload. This site earns a commission if you subscribe through that link, at no extra cost to you.