Instagram Reel Transcript

Give it a list of Instagram reels. Get back the spoken words out of each one: punctuated, split into paragraphs, with the language detected automatically.

You are never charged for a reel we could not fetch. On an ordinary list that is about one reel in four: private posts, deleted posts, and posts Instagram shows as a still image only. Those rows are written to your dataset so your list stays complete, and they cost you nothing.

What you get

• The transcript, as one block of text and again as readable paragraphs.
• Languages heard in each reel, comma separated. A reel that switches between two languages reports both, because the model labels the switch rather than picking a winner.
• Audio length and word count on the same row, so you can spot a 3-second clip and a 3-minute monologue without opening either.
• Confidence from the model, so you can filter out the noisy ones.
• A record of how the audio was reached, which is the first thing you want when a row comes back empty.

What it is for

• Content research: read a competitor's hooks instead of watching 200 reels.
• Repurposing: turn your own reels into captions, blog drafts or newsletter copy.
• Ad research: pull the script out of creatives that are working.
• Search: make a library of reels actually searchable by what is said in them.

Input

Paste reel links, one per line. A /reel/, /p/ or /tv/ link all work, and so does a bare shortcode.

{ "reels": ["https://www.instagram.com/reel/C8CaBfWs1mr/", "DKw2J6TMZd7"] }

The same reel shared in two different forms counts as one, so you are never charged twice for the same video.

Upload a file instead: .txt or .csv, one link per line. In a CSV the first column that looks like a reel is taken.

A link to a remote list works too: {"reels": [{"requestsFromUrl": "https://example.com/reels.txt"}]}.

The field can also be called urls, startUrls, videos, links or url.

Options

Option	What it does
language	Leave as multi to detect and switch languages automatically, which is what reels need. Set a code (en, es, de, fr, pt, hi) only when the whole list is in one language.
maxConcurrency	Leave empty. The Actor takes 4 per CPU core, and Apify gives one core per 4 GB of run memory.

Output

One row per reel.

{
  "input": "https://www.instagram.com/reel/C8CaBfWs1mr/",
  "url": "https://www.instagram.com/reel/C8CaBfWs1mr/",
  "shortcode": "C8CaBfWs1mr",
  "status": "ok",
  "transcript": "When the game you love loves you back, that is priceless.",
  "transcriptFormatted": "When the game you love loves you back,\n\nthat is priceless.",
  "wordCount": 11,
  "durationSeconds": 41.3,
  "language": "es, en",
  "confidence": 0.98,
  "meta": { "fetchedAt": "2026-09-20T09:12:03.114Z", "source": "deepgram-nova-3", "method": "url", "durationMs": 4820 }
}

status is ok, no-speech or unavailable.

Status	What it means	Charged
ok	Speech was found and transcribed.	yes
no-speech	The audio was fetched and processed, and there are no words in it. A music-only reel is the usual case.	yes
unavailable	The reel could not be reached at all: private, deleted, or Instagram shows only a still image for it.	no

Why no-speech is charged. You cannot know in advance whether a reel has anyone speaking in it - that is part of what you are asking. "Nobody speaks in this one" is an answer, and it costs exactly as much to produce as a transcript: ninety seconds of music is ninety seconds of audio. What you are never charged for is a reel we could not get, because then we have no answer for you at all.

Notes

• Not every reel can be reached, and you are not charged for the ones that cannot. Measured on 29 reels across two samples: the 14 most-viewed reels on Instagram, and 15 ordinary ones taken from recipe, fitness and tutorial posts. 23 of 29 returned video. The rest show a still image only, which is what Instagram does for private posts, deleted posts and some music posts. On the ordinary sample, of 15 reels submitted: 10 came back with a transcript, 1 was music with no speech, 4 could not be reached. You paid for 11 of the 15.
• A reel with no speech is not a failure. Plenty of reels are music and captions. You get a row saying so, with the audio length on it, so you can tell a silent reel from one we never reached.
• Length varies a lot, and it does not change your price. Across the ordinary sample the median reel ran 60 seconds and the longest 125. You pay per reel, not per minute.
• Transcripts are a model's best guess. Slang, names and crosstalk are where they go wrong. confidence is on every row for exactly that reason.