Instagram Reels, TikTok & Facebook Transcript by Username — Video to Text

Type a username — @zachking, nasa, or a profile link — and get the transcript of that creator's newest Instagram Reels or TikToks: what is actually said, the hook (first line), timestamps, language, views and likes. Or paste single TikTok, Instagram Reel and Facebook Reel links. One Actor for all three platforms, no login, no cookies.

Profile	Hook (first ~3 s)	Transcript (start)	Views
@nasa (Instagram)	Think about hundreds of square miles where we have different assets.	Think about hundreds of square miles where we have different assets. Definitely mobility systems…	3.6M
@nasa (TikTok)	NASA is accelerating Artemis.	NASA is accelerating Artemis. In this week, we shared some major updates on our path back to the moon…	163K
@mkbhd (TikTok)	Alright, so the memory price hikes have finally hit Apple and they have really hit.	Alright, so the memory price hikes have finally hit Apple and they have really hit. So already on Apple.com…	1.2M

(Real output of this Actor, 25 Oct 2026.)

{ "handles": ["@nasa", "tiktok:@nasa", "tiktok:@mkbhd"], "maxVideosPerProfile": 10 }

Speech recognition is OpenAI Whisper (99 languages, automatic language detection), running inside the Actor — no third-party transcription API, no extra keys. You only pay for videos that were transcribed — music-only clips, private profiles, deleted videos, login walls and duplicates come back as a status row and are free.

• Instagram Reels transcript by username — newest (or most viewed) Reels of any public profile, up to 500 per profile, no login.
• TikTok profile transcript — newest (or most viewed) videos of any public TikTok account, up to 500 per profile.
• Facebook Reel transcript — public Facebook Reels and videos, including fb.watch short links.
• Single links — TikTok, Instagram Reels, Facebook Reels / videos, short links (vm.tiktok.com, vt.tiktok.com, fb.watch).
• Hook line — the opening sentence of every video as its own column, with hookStartSeconds / hookEndSeconds, plus speaking pace (wordsPerMinute).
• Filters — only posts newer than a date or "7 days", skip pinned videos, minimum views, newest vs most viewed. Filtered videos are never downloaded or charged.
• Outliers — viewsVsCreatorMedian and isOutlier show which videos beat the creator's usual views.
• Timestamps, SRT, WebVTT, per-word timing — and chunks of N words for RAG / embeddings.
• English translation — translationEn for non-English videos, by Whisper itself, same price.
• Top TikTok comments — most-liked comments (pinned first) next to the transcript.
• Chain from any scraper — give a datasetId from an Instagram / TikTok / Facebook scraper run and every video in it is transcribed.
• Watchlists — named monitoring jobs; every row says isNew and firstSeenAt, and Only new videos skips what you already have.
• Honest billing — a charged column on every row, so your bill reconciles to the row.

For AI agents & LLMs — copy this block

Actor:   tactful_anvil/short-video-transcript  (Apify Store)
Does:    Instagram / TikTok username -> that creator's newest videos -> spoken transcript,
         hook (opening line), language, timestamped segments, optional SRT, views/likes/comments.
         Also single TikTok, Instagram Reel and Facebook Reel/video links. No login, no cookies.
Run:     POST https://api.apify.com/v2/acts/tactful_anvil~short-video-transcript/run-sync-get-dataset-items?token=APIFY_TOKEN
         (sync waits up to 5 minutes; for big batches use /runs and read the dataset afterwards)
Input    (every field optional; give handles and/or urls):
  {
    "handles": ["@nasa", "tiktok:@mkbhd", "https://www.instagram.com/zachking/"],
                                  // @user / user / profile URL; prefix tiktok: or ig: picks the platform per line
    "profilePlatform": "instagram",   // platform for plain usernames: instagram | tiktok | both
    "maxVideosPerProfile": 10,    // videos per profile after filters, up to 500 (Instagram + TikTok)
    "onlyPostsNewerThan": "",     // "2026-10-01" or "7 days" / "24 hours" / "2 weeks" / "3 months"
    "skipPinned": false,          // leave out pinned videos
    "minViews": 0,                // only videos with at least this many views
    "sortBy": "newest",           // newest | most_viewed (top by views among ~3x the videos listed)
    "urls": ["https://www.tiktok.com/@nasa/video/7699974130153590029",
             "https://www.instagram.com/reel/DeuH0dDFGcg/",
             "https://www.facebook.com/reel/1795689255099117"],
    "datasetId": "",              // dataset of any Instagram / TikTok / Facebook scraper run -> transcribe its videos
    "datasetItems": [],           // or paste scraper rows directly (any shape with a url / webVideoUrl / shortCode)
    "onlyNew": false,             // true: skip videos an earlier run of yours already transcribed
    "watchlistId": "",            // named memory per monitoring job; rows get isNew + firstSeenAt
    "language": "",               // ISO 639-1 like "en", "es", "id"; empty = auto-detect per video
    "model": "base",              // base (recommended) | tiny (fastest, clear English)
    "preferNativeCaptions": true, // use captions the platform already publishes (TikTok) when present
    "includeSegments": true,      // segments [{start,end,text}] in seconds
    "includeSrt": false,          // add an SRT subtitle string
    "includeVtt": false,          // add a WebVTT subtitle string
    "wordTimestamps": false,      // add words[{start,end,word}] inside every segment
    "chunkWords": 0,              // >0: add chunks[{index,start,end,text,wordCount}] of ~N words (RAG)
    "translateToEnglish": false,  // add translationEn (Whisper speech translation), same price
    "maxComments": 0,             // >0: add topComments (TikTok only; IG/FB need a login)
    "maxVideos": 100,             // hard cap on unique videos per run (links + all profiles), up to 5000
    "maxDurationSec": 180         // transcribe at most this many seconds of each video (10-600)
  }
  Empty input {} -> a FREE 2-Reel sample from Instagram @nasa (nothing charged but the start fee).
Output   one row per video (or per profile that could not be listed):
         url, platform, profile, status, charged, error, videoId, author, authorId, caption,
         durationSec, uploadedAt, viewCount, likeCount, commentCount, shareCount, saveCount,
         musicTitle, musicArtist, thumbnailUrl, ownerFullName, isVerified, followerCount,
         isPinned, isPaidPartnership, taggedUsers[], coauthors[], hashtags[], mentions[],
         hook, hookStartSeconds, hookEndSeconds, hasSpeech, transcript, wordCount,
         wordsPerMinute, language, languageProbability, transcriptSource (whisper |
         native_captions), model, speechSeconds, truncated, segments[], processedAt,
         creatorMedianViews, viewsVsCreatorMedian, isOutlier
         + when asked: srt, vtt, chunks[], translationEn, topComments[], isNew, firstSeenAt.
         Run receipt: key-value store record SUMMARY (counts per status, charged, filtered,
         per-profile listed/selected) — reconciles with the per-row `charged` column.
         status "ok" = transcribed and charged. Any other status is FREE and `error` says why:
         no_speech, no_videos, caption_only, blocked, network_error, private, login_required,
         not_found, no_audio, geo_restricted, download_failed, transcribe_failed,
         budget_limit, timeout, invalid, unsupported, youtube.
         The run always ends SUCCEEDED; its status message sums up the rows per status.
Pricing  pay per event: $0.008 per transcribed video (up to 3 min of audio) + start fee $0.001 per GB
         of run memory ($0.002 with the default 2 GB).
         Platform usage is included. Respects "Maximum cost per run": videos that would exceed it
         come back as budget_limit rows and are not transcribed or charged.
Not for  YouTube (returns status youtube, free), private / followers-only / age-restricted videos.

MCP in one line: add https://mcp.apify.com/?tools=tactful_anvil/short-video-transcript as an MCP server in Claude, Cursor, VS Code or any MCP client (sign in with your Apify account or pass your token). The Actor then shows up as one tool your agent can call: "Transcribe the 10 newest Reels of @garyvee and list the hooks that start with a question."

What does this Actor do?

It turns short-form video into text, at scale, from three platforms:

1. Lists profiles. For every username you give, it fetches that creator's public videos — Instagram Reels through Instagram's public Reels tab (12 per page, paged up to 500), TikTok videos through the public profile page. Pinned videos are sorted back into date order (or skipped), then your filters pick which ones to transcribe: newer than a date, minimum views, newest or most viewed.
2. Gets the audio. For each video it downloads only the smallest file that carries the sound (Instagram profile listings already include the video file, so no extra page visit is needed).
3. Uses existing captions when they exist. Many TikToks ship captions published by the creator or by TikTok. When Use the platform's own captions is on, those are returned directly — faster, and exactly what TikTok shows.
4. Transcribes everything else with Whisper. The audio is converted to 16 kHz mono and transcribed by faster-whisper (OpenAI Whisper weights, int8) inside the Actor's own container. Voice activity detection skips music and silence, which also tells the Actor when a clip has no speech at all.
5. Writes one clean row per video — transcript, hook, segments, optional SRT, language, word count, speech seconds, plus the post's caption, views, likes, comments, shares, sound title, upload date and thumbnail.
6. Bills only real transcripts. A clip that is only music, a deleted link or a login wall becomes a free status row with a plain-language reason.

Two videos are processed at the same time (one downloading while the other is being transcribed), and every platform request is spaced out politely.

Who is this for?

• Content repurposers & clippers — turn viral short videos into scripts, hooks, captions, blog posts, newsletters and threads.
• Marketers & agencies — study what top creators and competitors actually say in their Reels and TikToks, not just what they write in the caption.
• UGC & ad researchers — collect ad scripts and opening hooks from trending videos into a swipe file.
• Social media managers — subtitle your own videos (SRT) and keep a searchable archive of everything your brand has said.
• Researchers & analysts — speech-level datasets from public short-form video, ready for LLMs, sentiment, topic modelling or claim checking.
• Brand & reputation monitoring — watch what creators say about a product, a launch or a category, every day.
• AI agents (MCP) — "transcribe these 20 TikToks and summarize the hooks" is one tool call.

Input

You can mix every kind of input in one run. All fields are optional; with nothing at all you get a free sample.

Field	Type	Default	What it does
handles	list of strings	—	Profiles to list: @zachking, zachking, https://www.instagram.com/zachking/, https://www.tiktok.com/@zachking. Prefix tiktok: or ig: (tiktok:@nasa, ig:nasa) to pick the platform per line.
profilePlatform	instagram | tiktok | both	instagram	Where plain usernames are looked up. both lists the same username on Instagram and TikTok. Profile links always use their own platform.
maxVideosPerProfile	integer 1–500	10	Videos to transcribe per profile, after the filters.
onlyPostsNewerThan	string	empty	2026-10-01, 7 days, 24 hours, 2 weeks, 3 months, 1 year. Older videos are skipped before download. For Instagram the listing stops paging as soon as a whole page is older.
skipPinned	boolean	false	Leave out videos pinned to the top of the profile.
minViews	integer	0	Only videos with at least this many views (videos with hidden views are kept).
sortBy	newest | most_viewed	newest	most_viewed lists about 3× Videos per profile (at least 36) and transcribes the most viewed of them.
urls	list of strings	—	Single videos: TikTok, Instagram Reels (/reel/…, /p/…), Facebook Reels and videos, short links. Profile links pasted here are listed like handles. Tracking parameters are removed and duplicates skipped.
datasetId	string	—	Dataset ID of another Actor run (Instagram / TikTok / Facebook scraper). Post links are read from url, webVideoUrl, postUrl, shortCode and similar fields; Instagram rows with a videoUrl file are transcribed straight from that file. Photo posts are skipped.
datasetItems	array	—	The same, with rows pasted directly.
onlyNew	boolean	false	Skip videos an earlier run of yours already finished (transcribed, or found to have no speech / no audio). Remembered in a key-value store in your own Apify account (short-video-transcript-seen, or your watchlist's store).
watchlistId	string	—	A named memory per monitoring job (competitors, client-acme) → store svt-watch-<name>. Rows get isNew and firstSeenAt.
language	string	empty	Spoken language as an ISO 639-1 code (en, es, pt, id, de, fr, hi, ar, ja, …). Empty = detected separately for every video. Setting it helps with very short clips and mixed-language creators.
model	base | tiny	base	base = good accuracy, recommended. tiny = fastest, fine for clear English speech. Same price.
preferNativeCaptions	boolean	true	Return captions the platform already publishes (mostly TikTok) instead of running speech recognition. Turn off to always transcribe the audio.
includeSegments	boolean	true	Add segments: [{start, end, text}] in seconds.
includeSrt	boolean	false	Add srt: a ready-to-save SubRip subtitle string.
includeVtt	boolean	false	Add vtt: a WebVTT subtitle string (HTML5 video, YouTube upload, most editors).
wordTimestamps	boolean	false	Add words [{start, end, word}] inside every segment. Always uses Whisper (not platform captions).
chunkWords	integer 0–2000	0	Add chunks: the transcript cut into pieces of about N words with start/end seconds, for embeddings / RAG.
translateToEnglish	boolean	false	Add translationEn: the speech translated to English by Whisper (non-English videos; English videos get their transcript). Same price.
maxComments	integer 0–50	0	Add topComments for every transcribed TikTok: most liked first, pinned on top. Instagram and Facebook hide comments from logged-out visitors → null.
maxVideos	integer 1–5000	100	Stop after this many unique videos in total (links + all profiles). Extra profiles/links are counted in the status message, not charged.
maxDurationSec	integer 10–600	180	Transcribe at most this many seconds of each video. Longer videos are transcribed from the start and marked truncated: true.
proxyConfiguration	proxy	off	Not needed for most runs. If a platform keeps answering blocked, a residential proxy can help.

Input examples

Instagram Reels transcript by username (the 12 newest Reels of three creators):

{ "handles": ["@garyvee", "@hormozi", "nasa"], "profilePlatform": "instagram", "maxVideosPerProfile": 12 }

TikTok profile transcript (50 newest TikToks of one account, with SRT):

{ "handles": ["tiktok:@mkbhd"], "maxVideosPerProfile": 50, "includeSrt": true }

Same creator on both platforms:

{ "handles": ["@nasa"], "profilePlatform": "both", "maxVideosPerProfile": 5 }

Facebook Reel transcript & single links:

{
  "urls": [
    "https://www.facebook.com/reel/1795689255099117",
    "https://fb.watch/xyz123/",
    "https://vm.tiktok.com/ZMabc123/",
    "https://www.instagram.com/reel/DeuH0dDFGcg/"
  ]
}

Spanish creators, daily, only new videos:

{ "handles": ["tiktok:@creador1", "tiktok:@creador2"], "language": "es", "maxVideosPerProfile": 10, "onlyNew": true }

Output

One row per video. Example (real row, segments shortened):

{
  "url": "https://www.instagram.com/reel/DeuH0dDFGcg/",
  "platform": "instagram",
  "profile": "@nasa",
  "status": "ok",
  "charged": true,
  "error": null,
  "videoId": "DeuH0dDFGcg",
  "author": "nasa",
  "authorId": "528817151",
  "caption": "There's a lot of ground to cover when you're building a Moon Base. …",
  "durationSec": 45.82,
  "uploadedAt": "2026-10-20T15:10:43.000Z",
  "viewCount": 3671149,
  "likeCount": 54014,
  "commentCount": 447,
  "shareCount": null,
  "musicTitle": "Original audio",
  "thumbnailUrl": "https://scontent-iad6-1.cdninstagram.com/…",
  "hook": "Think about hundreds of square miles where we have different assets.",
  "transcript": "Think about hundreds of square miles where we have different assets. Definitely mobility systems. There's a lot of people spend a lot of time making sure every detail is right …",
  "wordCount": 141,
  "language": "en",
  "languageProbability": 0.98,
  "transcriptSource": "whisper",
  "model": "whisper-base",
  "speechSeconds": 43.15,
  "truncated": false,
  "segments": [
    { "start": 0, "end": 5.68, "text": "Think about hundreds of square miles where we have different assets. Definitely" },
    { "start": 5.68, "end": 10.04, "text": "mobility systems. There's a lot of people spend a lot of time making sure" },
    { "start": 10.04, "end": 13.88, "text": "every detail is right and it's really specific to the job it needs to do." }
  ],
  "processedAt": "2026-10-25T04:30:09.093Z"
}

A music-only Reel from the same run — free, and it says why:

{
  "url": "https://www.instagram.com/reel/Dez3pRXgW1d/",
  "platform": "instagram",
  "profile": "@nasa",
  "status": "no_speech",
  "charged": false,
  "error": "No spoken words found (music, sound only or silence) — not charged.",
  "durationSec": 32.17
}

All output fields

Field	Example	Notes
url	https://www.tiktok.com/@nasa/video/7699974130153590029	the video link without tracking parameters (for a profile that could not be listed: the profile link)
platform	tiktok	tiktok, instagram, facebook — or invalid / unsupported / youtube for entries this Actor does not handle
profile	@nasa	the username this video was listed from (null for videos given as links)
status	ok	see the status table below; only ok is charged
charged	true	true only for status ok — sum this column to check your bill
error	null	plain-language reason for every status other than ok
videoId	7699974130153590029	platform id (Instagram shortcode, TikTok id, Facebook id)
author, authorId	nasa, 528817151	username / page name and account id
caption	There's a lot of ground to cover…	what the creator wrote — not what was said
durationSec	45.82	video length
uploadedAt	2026-10-20T15:10:43.000Z	ISO timestamp of the post
viewCount	3671149	views / plays when shown publicly
likeCount	54014	likes (Facebook: reactions)
commentCount, shareCount	447, null	when the platform shows them publicly (Instagram hides shares)
saveCount	1498	saves / bookmarks (TikTok)
musicTitle, musicArtist	Original audio, nasa	sound / track title and its artist
ownerFullName, isVerified, followerCount	NASA, true, 104378961	creator display name, badge, followers (followers: the listed Instagram profile's own Reels)
isPinned	false	pinned to the top of the profile
isPaidPartnership, taggedUsers, coauthors	false, ["esa"], ["nasaadmin"]	Instagram sponsorship label, tagged accounts, collab co-authors
hashtags, mentions	["ArtemisII"], ["esa"]	from the caption, without # / @
thumbnailUrl	https://…	cover image (platform links expire after a few days — save it if you need it)
hook	NASA is accelerating Artemis.	the opening line: speech that starts in the first ~3 seconds, cut at the first full sentence, max 20 words
hookStartSeconds, hookEndSeconds	0.11, 2.13	where the hook sits in the video
hasSpeech	true	false for music-only / silent clips (no_speech)
transcript	NASA is accelerating Artemis. In this week…	everything said, as one text
wordCount	141	spoken words (sound tags like "[Music]" excluded)
wordsPerMinute	199	speaking pace over the seconds that contain speech
language, languageProbability	en, 0.98	detected (or your requested) language and the detector's confidence
transcriptSource	whisper	whisper (speech recognition) or native_captions (captions published on the platform)
model	whisper-base	speech model used (null for native captions)
speechSeconds	43.15	seconds of actual speech found; music and silence excluded
truncated	false	true when the video was longer than Max video length and only its start was transcribed
segments	[{"start":0,"end":5.68,"text":"…"}]	timestamped pieces in seconds (when Timestamped segments is on)
srt	1\n00:00:00,000 --> 00:00:05,680\n…	SRT subtitle text (when SRT subtitles is on)
vtt	WEBVTT\n\n00:00:00.000 --> 00:00:05.680\n…	WebVTT subtitle text (when WebVTT subtitles is on)
segments[].words	[{"start":0,"end":0.4,"word":"Alright"}]	per-word timing (when Per-word timestamps is on)
chunks	[{"index":0,"start":0,"end":12.17,"wordCount":32,"text":"…"}]	~N-word pieces for RAG (when Chunks > 0)
translationEn	They gave me a…	English translation (when Translate to English is on)
topComments	[{"text":"Hi NASA, when is Artemis III launching?","likes":13,"author":"…","pinnedByAuthor":false,…}]	TikTok only (when Top comments > 0)
creatorMedianViews	5139534	median views of the videos listed from this profile (5+ needed)
viewsVsCreatorMedian, isOutlier	21.12, true	this video's views ÷ the creator's median; outlier = 2× or more
isNew, firstSeenAt	true, 2026-10-26T06:16:49Z	with a watchlist or Only new videos: first time seen and when
processedAt	2026-10-25T04:30:09.093Z	when the row was produced

Every run also writes a receipt to its key-value store, record SUMMARY: rows per status, how many were charged, how many were filtered out (older / pinned / views), duplicates, and per profile how many videos were listed and selected.

Download the dataset as JSON, CSV, Excel, XML, HTML or RSS, or read it through the API. The Transcripts view in the Console shows profile, status, hook, transcript, views, likes, length, language, author, caption and link side by side.

Status values — what every row means

Status	Charged	Meaning	What to do
ok	yes	Transcribed.	—
no_speech	no	Only music, sound effects or silence — nothing was said.	Nothing; this is the correct answer for that clip.
no_videos	no	The profile exists but shows no public videos (or is private).	Check the username.
caption_only	no	Instagram did not serve this single Reel's file without a login; the public caption is returned instead.	Use the creator's username in Profiles — profile listings usually include the file.
blocked	no	The platform refused the request (temporary IP block; on TikTok also deleted / non-existent videos).	Open the link in a browser. If it works, retry later or turn on a residential proxy.
network_error	no	The connection dropped twice.	Retry later.
private	no	Private or followers-only.	Never accessed by design.
login_required	no	The platform wants a login for this video.	Never bypassed by design.
not_found	no	Deleted, never existed, or a mistyped link.	Check the link.
no_audio	no	The video has no sound track.	—
geo_restricted	no	Not available from the region the run connected from.	A proxy in the right country can help.
download_failed	no	The file could not be downloaded for another reason.	Retry; report the link if it keeps failing.
transcribe_failed	no	The audio file could not be decoded.	Report the link.
budget_limit	no	Not processed because the run reached your Maximum cost per run.	Raise the limit or split the input.
timeout	no	Not processed because the run was about to hit its timeout.	Raise the run timeout (Run options) or lower Max videos.
invalid	no	Not a URL or username.	Fix the line.
unsupported	no	A link from another site.	—
youtube	no	YouTube is not handled by this Actor.	—

Every run ends SUCCEEDED and its status message sums it up, for example: "7 transcribed (charged) · 2 no speech · 1 not found (free) · 3 duplicate video(s) skipped". A bad link never fails the run.

Pricing — pay only for transcripts

• $0.008 per transcribed video (up to 3 minutes of audio) — $8 per 1,000 transcripts.
• Start fee: $0.001 per GB of run memory — $0.002 per run with the default 2 GB.
• Platform usage is included — no compute, proxy or storage costs on top.
• Free: no_speech, private, not_found, login_required, blocked, no_audio, caption_only, duplicates, links over your limits, and anything not processed because of your cost limit or the timeout.

Job	Transcribed videos	Cost
Try it: empty input (2-Reel sample)	2 (free)	$0.002
1 creator, 12 newest Reels	~10 (2 are often music-only → free)	~$0.08
10 competitors × 10 TikToks, once	~90	~$0.72
Daily monitor, 20 creators, Only new videos on (~1 new video per creator per day)	~20 / day	~$0.17 / day · ~$5 / month
Hook library: 1,000 videos	~900	~$7.20

Maximum cost per run is respected exactly: before transcribing a video, the Actor checks that the transcript still fits in your limit. Videos that would go over it come back as budget_limit rows (free) instead of being transcribed and thrown away — you never pay for work you cannot receive.

How to use it (no code)

1. Open the Input tab.
2. Profiles: type usernames or profile links, one per line. Plain usernames use Platform for usernames (Instagram by default, or TikTok, or both); tiktok:@name / ig:name choose per line.
3. Videos per profile: how many of the newest videos to transcribe (default 10).
4. Or paste single links into Video URLs (short links like vm.tiktok.com/… and fb.watch/… work). Profiles and links can be mixed.
5. Optional: set the Spoken language (en, es, id, pt, …) if you already know it.
6. Click Start. Rows appear in the Output tab while the run is going. Export as CSV / Excel / JSON.

Started with an empty input, the Actor transcribes a free 2-Reel sample from Instagram @nasa, so you can see the output before spending anything.

Recipes

1. Daily creator monitoring (only new videos)

Watch competitors or favourite creators and get each new video's transcript the morning after it is posted.

1. Input: { "handles": ["@creator1", "@creator2", "tiktok:@creator3"], "maxVideosPerProfile": 12, "onlyNew": true }.
2. Save it as a Task, then Schedule it daily (Console → Schedules → Create, e.g. 0 7 * * *).
3. The first run transcribes the backlog (up to Videos per profile each); every later run transcribes only what was posted since. Already-seen videos are skipped before download, so they cost nothing.
4. Connect the dataset to Google Sheets, Slack, email or a webhook (see Integrations) to get the new transcripts delivered.

The memory lives in a key-value store named short-video-transcript-seen in your own account. Delete that store to start fresh. Music-only clips are remembered too, so they are not re-checked every day.

1b. One memory per client: watchlists

Running monitors for several clients or niches? Give each schedule its own watchlistId ("client-acme", "fitness-creators"). Each gets a separate memory (svt-watch-client-acme), and every row carries isNew and firstSeenAt — even without Only new videos, so you can flag fresh videos in a full refresh.

2. Hook swipe file

Collect the opening lines of what works in your niche:

1. Input: 10–30 top creators, maxVideosPerProfile: 12.
2. Export to Excel / Google Sheets and sort by viewCount.
3. Read the hook column next to viewCount and likeCount — the hooks of the top 10% of videos are your swipe file.

hook is built from the real timestamps (speech that starts in the first ~3 seconds), so it is what the viewer actually hears first — not the caption. Faster: { "handles": [...], "sortBy": "most_viewed", "maxVideosPerProfile": 10, "onlyPostsNewerThan": "90 days" } transcribes only each creator's top 10 of the last 90 days.

2b. Find outliers — what beat the creator's usual

Every video listed from a profile is compared with that creator's median views: viewsVsCreatorMedian: 4.2 means 4.2× the usual, isOutlier: true from 2×. Sort by it to see which topics and hooks broke out — a better signal than raw views, which favour big accounts.

3. Repurpose videos into blog posts, threads and newsletters

Feed transcript + caption of each row to ChatGPT / Claude with a prompt like "Rewrite this transcript as a 600-word blog post with H2 headings, keep the speaker's examples." With the API, one Make/Zapier/n8n scenario does it for every new row.

4. Subtitles for your own videos

Turn on SRT subtitles and paste your own Reel / TikTok / Facebook links. Save the srt field as video.srt and upload it to your editor (CapCut, Premiere, DaVinci Resolve) or to the platform. segments gives the same timing as JSON for burned-in captions.

5. Multilingual research

Leave Spoken language empty and every video gets its own detected language and languageProbability. Filter the dataset by language, or set language when a creator always speaks the same one (more accurate on very short clips). Whisper covers 99 languages, including English, Spanish, Portuguese, Indonesian, Hindi, Arabic, German, French, Japanese, Korean, Turkish, Vietnamese and Russian.

6. Chain after any scraper (datasetId)

Ran an Instagram, TikTok or Facebook scraper already? Pass its dataset: { "datasetId": "<dataset id>" }. The Actor reads the post link from the usual fields (url, webVideoUrl, postUrl, shortCode, …), skips photo posts and duplicates, and — for Instagram rows that include the videoUrl file — transcribes straight from that file with the row's own views, likes and caption. Video file links from scrapers expire after a few hours, so run it soon after the scraper (a webhook or Make/Zapier step does that). Links from a spreadsheet work too: paste them into urls (up to 5,000 per run with maxVideos).

7. Transcripts for AI / RAG

Set chunkWords to 150–300: each row gets chunks with start/end seconds, ready to embed. Store the video url + chunk.start as metadata and your chatbot can cite the exact second. wordTimestamps: true adds per-word timing for word-level search or karaoke captions.

8. Foreign-language creators in English

{ "handles": ["tiktok:@luisitocomunica"], "translateToEnglish": true } gives the Spanish transcript and an English translationEn side by side. Whisper's translation is literal — good for research and search; polish it with an LLM before publishing.

Use it via API

One call, wait for the rows (best for up to ~20 videos):

curl -X POST "https://api.apify.com/v2/acts/tactful_anvil~short-video-transcript/run-sync-get-dataset-items?token=$APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{ "handles": ["tiktok:@nasa"], "maxVideosPerProfile": 3 }'

Python (pip install apify-client):

from apify_client import ApifyClient

client = ApifyClient("APIFY_TOKEN")
run = client.actor("tactful_anvil/short-video-transcript").call(run_input={
    "handles": ["@nasa", "tiktok:@mkbhd"],
    "maxVideosPerProfile": 10,
    "onlyNew": True,
})
for row in client.dataset(run["defaultDatasetId"]).iterate_items():
    if row["status"] == "ok":
        print(row["profile"], row["viewCount"], row["hook"])

Node.js (npm i apify-client):

import { ApifyClient } from 'apify-client';

const client = new ApifyClient({ token: process.env.APIFY_TOKEN });
const run = await client.actor('tactful_anvil/short-video-transcript').call({
  urls: ['https://www.tiktok.com/@nasa/video/7699974130153590029'],
  includeSrt: true,
});
const { items } = await client.dataset(run.defaultDatasetId).listItems();
console.log(items[0].transcript);

Big batches: start with POST /v2/acts/tactful_anvil~short-video-transcript/runs, then poll the run or use a webhook on ACTOR.RUN.SUCCEEDED, and read GET /v2/datasets/{defaultDatasetId}/items?format=csv.

Use with AI agents (MCP)

This Actor is available through the Apify MCP server. To give your agent exactly this one tool:

{
  "mcpServers": {
    "short-video-transcript": {
      "url": "https://mcp.apify.com/?tools=tactful_anvil/short-video-transcript"
    }
  }
}

Works with Claude (Desktop, Code, claude.ai connectors), ChatGPT, Cursor, VS Code, Windsurf, n8n's MCP client and the OpenAI Agents SDK. Prompts that work well:

• "Transcribe the 12 newest Reels of @hormozi and give me the 5 strongest hooks with their view counts."
• "Here are 15 TikTok links — transcribe them and tell me which ones mention our brand."
• "Every morning, check @competitor1 and @competitor2 on TikTok for new videos and summarize what they said." (with onlyNew: true)

Why it is agent-friendly: flat rows, a status on every row instead of exceptions, charged for cost tracking, hook pre-extracted, and the run never fails on a bad link — the agent gets a readable reason and can carry on.

Integrations

• Google Sheets — Apify's Google Sheets integration (or Make / Zapier) appends every new transcript as a row.
• Make — "Apify → Watch Actor Runs" → "Get Dataset Items" → any of 1,500+ apps.
• Zapier — "New Finished Actor Run" trigger → filter status = ok → Notion, Airtable, Slack, Gmail.
• n8n — the Apify node (run Actor + get dataset items), or the MCP client node with the URL above.
• Webhooks — POST to your endpoint on every finished run.
• LangChain / LlamaIndex — load the dataset with the Apify loaders and index transcript with segments for timestamped retrieval.

Instagram Reels transcript by username

Type the username (@garyvee, garyvee or https://www.instagram.com/garyvee/) into Profiles and keep Platform for usernames on Instagram Reels. The Actor reads the profile's Reels tab the way a logged-out visitor sees it and transcribes the newest Reels — up to 500 per profile, read 12 per page with polite pauses, never logged in. Pinned Reels are put back into date order or skipped (skipPinned). Collab Reels that appear on the profile are included with their real owner in author. Private accounts return private (free). Views, likes, comments, caption, sound and upload date come with every row.

TikTok profile transcript

Use tiktok:@username, a https://www.tiktok.com/@username link, or set Platform for usernames to TikTok. Up to 500 videos per profile, newest or most viewed. Many TikToks carry captions published by the creator or TikTok — with Use the platform's own captions on, those are returned directly (transcriptSource: native_captions); otherwise the audio is transcribed with Whisper. If TikTok refuses to list a profile to cloud servers, that profile comes back as blocked (free) after one automatic retry; single video links are not affected.

Facebook Reel transcript

Paste public Facebook Reels or video links into Video URLs — facebook.com/reel/…, facebook.com/watch?v=…, facebook.com/<page>/videos/…, facebook.com/share/r/… and fb.watch/… short links. You get the transcript plus views and reactions (parsed from the public page), page name and caption. Facebook profiles by username are not listed; use links.

TikTok / Instagram video to text — how it compares

• Spoken words, not just captions. Many tools return the caption the creator typed. This Actor returns what is said, with timestamps, and tells you which source it used.
• Three platforms, one schema. TikTok, Instagram and Facebook rows have the same columns, so one sheet or one prompt handles all of them.
• Usernames, not just links. Give a creator, get their newest videos.
• You pay for transcripts, not attempts. Music-only clips are detected by voice activity detection and are free; so is everything that failed.
• No login, no cookies, no third-party API key. Whisper runs inside the Actor.
• Built for schedules. Only new videos and named watchlists skip already-processed videos before download, so a daily monitor costs only what is new.
• Filters before download. Date, pinned, views and most-viewed are applied before any file is fetched — nothing filtered is charged.
• More than a transcript. Hook timing, speaking pace, outlier score, hashtags, collaborators, English translation, TikTok comments, chunks for RAG — in the same row, same price.

Platform notes (honest)

• TikTok — public videos: transcribed. Creator / TikTok captions are used when present (faster).
• TikTok profiles — TikTok sometimes refuses to list a profile to cloud servers; that profile then comes back as blocked (free). Single TikTok video links are not affected — paste them into Video URLs, or retry with a proxy.
• Facebook — public Reels and videos: transcribed.
• Instagram profiles — up to 500 Reels of public profiles, listed 12 per page. Private profiles return private (free). If Instagram stops answering mid-listing, the Reels listed so far are still transcribed.
• Instagram single Reel links — transcribed when Instagram serves the file without a login; otherwise you get the public caption with status caption_only (free). Tip: the profile's username usually works better than single links.
• Photo posts and carousels have nothing to transcribe and are skipped by profile listings.
• YouTube is not handled here.
• Private, followers-only and age-restricted videos are never accessed.

Limits

• Instagram: up to 500 Reels per profile (12 per page, ~1.5 s per page).
• TikTok: up to 500 videos per profile; TikTok sometimes refuses profile listings to cloud servers (blocked, free).
• Comments: TikTok only. Follower counts: Instagram only.
• Up to 5,000 unique videos per run (maxVideos); split very large jobs into several runs.
• Up to 600 seconds of audio per video (maxDurationSec, default 180). Reels and TikToks are almost always shorter.
• Speed: about 5 typical 30–60 s Reels per minute with the default 2 GB memory (cloud test, 25 Oct 2026); long videos take longer. More memory = more CPU = faster speech recognition (4 GB gives a full CPU core).

FAQ

How do I get the transcript of an Instagram Reel? Paste the creator's username into Profiles (best) or the Reel link into Video URLs, then click Start. The transcript column has the spoken text, segments the timestamps.

How do I transcribe all TikToks of a profile? { "handles": ["tiktok:@username"], "maxVideosPerProfile": 500 }. Raise Max videos if you add more profiles.

Can I get a transcript of a Facebook Reel? Yes — paste the public Reel link (or fb.watch link) into Video URLs.

How accurate is it? Whisper base handles clear speech in major languages very well; heavy background music, slang and crosstalk lower accuracy. Set Spoken language for very short clips. Choose Fast (tiny) only for clear English speech.

What if a video is only music? You get a no_speech row and pay nothing. The voice activity detector measures real speech (speechSeconds), so a song with a few words in it is still transcribed when words are actually sung or spoken.

Does it translate? Yes, to English: turn on Translate to English and non-English videos get translationEn (Whisper's own speech translation, same price). Other target languages: translate transcript or translationEn with your LLM of choice.

Can I get only recent videos? "onlyPostsNewerThan": "7 days" (or a date). Older videos are skipped before download and cost nothing.

Can I get a creator's most viewed videos? "sortBy": "most_viewed" — the top videos by views among about 3× the number you ask for (at least 36). Add minViews for a hard floor.

Does it get comments? TikTok yes (maxComments, up to 50 most liked, pinned first). Instagram and Facebook do not show comments to logged-out visitors, and this Actor never logs in.

Does it read text shown on screen? Not yet — only spoken words (and creator captions). The written caption of the post is always included.

Does it work with private accounts? No. Only public videos, without logging in.

Why is shareCount empty for Instagram? Instagram does not show share counts to logged-out visitors.

Do thumbnail links expire? Yes, Instagram and TikTok image links expire after a few days. Download them right after the run if you need them.

Daily monitoring? Turn on Only new videos and schedule the run (Recipe 1). Videos already transcribed by your earlier runs are skipped and never charged twice.

Long videos? Audio after Max video length (default 180 s, max 600 s) is not transcribed and the row is marked truncated.

Can I limit my spend? Yes — set Maximum cost per run in Run options. The Actor stops transcribing before it would cross that limit and marks the rest budget_limit (free).

Is there an API / can my AI agent use it? Yes — see Use it via API and Use with AI agents (MCP) above.

Is this legal? The Actor only reads publicly available videos, without logging in, at a polite pace. You are responsible for how you use the transcripts (copyright, privacy, platform terms). Do not use it to collect personal data you have no legitimate reason to process.

Something broke? Open an Issue on this page with the failing link and your run ID — fixes usually ship within a day.

Changelog

• 0.3 (26 Oct 2026) — Instagram up to 500 Reels per profile (paging); filters (onlyPostsNewerThan, skipPinned, minViews, sortBy); datasetId / datasetItems input; named watchlists with isNew / firstSeenAt; hookStartSeconds / hookEndSeconds, wordsPerMinute, hasSpeech; WebVTT, per-word timestamps, chunks; owner name, verified, followers, pinned, paid partnership, tagged users, collaborators, hashtags, mentions, saves, sound artist; outlier score vs creator median; translationEn; TikTok top comments; SUMMARY receipt; two speech workers on 16 GB+ runs.
• 0.2 (25 Oct 2026) — Profiles by username (Instagram + TikTok), hook column, Only new videos, free sample on empty input, TikTok profile retry.
• 0.1 (24 Oct 2026) — First release: TikTok, Instagram Reels and Facebook Reels links → Whisper transcript, native TikTok captions, SRT, pay only for transcripts.