X Video Subtitle & Caption Transcript Converter

Public X (Twitter) videos either already carry captions or need speech-to-text to become searchable text. This actor turns either case into a clean, timestamped transcript — reusing the video's own captions when they exist, and running AI speech-to-text when they don't — and can hand back ready-to-use SRT or WebVTT subtitle files, not just raw text. No login or cookies are needed.

What this actor does

• Captions first: uses the video's own captions when the uploader or X provides them
• Whisper AI fallback: automatically transcribes videos without captions, in any language
• Three granularities: raw segments, sentences, or real per-word timings when available
• Five output formats: timestamped segments, full text, both, SRT or WebVTT — with SRT/VTT files saved for download
• Explicit outcomes: every post produces a row; posts with no transcript get transcriptAvailable: false and a clear message
• Empty fields are omitted

Output per transcript

Each row has a type: video transcript when a transcript was produced, otherwise status.

Field	Description
tweetUrl, tweetId	Post URL and ID
authorUsername, authorName, authorFollowers	Post author details (when Include post details is on)
tweetText, publishedAt	Post text and publish time (when Include post details is on)
engagementRate	Derived engagement in percent (when Include post details is on)
language	Detected or requested language code
transcriptionMethod	captions, ai or none
transcriptAvailable	true when a transcript was produced
modelUsed	Whisper model actually used for AI transcripts
modelNote	Explanation when a smaller model was substituted for memory reasons
segmentGranularity	segment, sentence or word — granularity actually applied
granularityRequested	Present when the requested granularity could not be applied
granularityNote	Explanation when a granularity fell back to raw segments
segmentCount	Number of timed segments
segments	[{index, start, end, text}]; includes real words[{word, start, end}] timings when available
fullTranscript	Joined transcript text
srt, srtUrl	SRT subtitle text and a signed download link for the .srt file
vtt, vttUrl	WebVTT subtitle text and a signed download link for the .vtt file
message	Explanation when no transcript could be produced
collectedAt	UTC timestamp

Unavailable transcripts are status rows with transcriptAvailable: false, transcriptionMethod: "none" and a message explaining the reason. When the failure happens while resolving a profile input (instead of a post), the row carries profileUrl (and profileUsername, when known) instead of tweetUrl/tweetId.

Input

Field	Type	Default	Description
postUrls	array	–	Required. Post links (x.com or twitter.com) that contain a video or GIF. Profile links or usernames are accepted too — the actor transcribes the profile's latest video post
profileUrls	array	–	Additional profiles to transcribe (latest video post each)
transcriptionMethod	string	auto	auto (captions first, AI fallback), captions (skip videos without captions), ai (always Whisper)
whisperModel	string	base	tiny, base, small, medium, large-v2
language	string	–	Optional spoken-language hint for AI (e.g. en, es, ur); ignored for native captions
outputFormat	string	segments	segments, fullText, both, srt, vtt
segmentGranularity	string	segment	segment (raw cues), sentence, word (only when real word timings exist)
includeWordTimestamps	boolean	false	Add real per-word start/end times to AI segments
includeTweetMetadata	boolean	true	Attach author, post text and publish date to each row
proxyConfiguration	object	–	Optional, for large or repeated runs

Example: captions first, AI fallback

{
  "postUrls": ["https://x.com/NASA/status/2107095768821576019"],
  "transcriptionMethod": "auto",
  "outputFormat": "both"
}

Example: always use AI with an English hint

{
  "postUrls": [
    "https://x.com/elonmusk/status/1519480761749016577",
    "NASA"
  ],
  "transcriptionMethod": "ai",
  "whisperModel": "small",
  "language": "en"
}

Example: SRT subtitle file

{
  "postUrls": ["https://x.com/NASA/status/2107095768821576019"],
  "outputFormat": "srt",
  "segmentGranularity": "sentence"
}

Example: word-level timestamps

{
  "postUrls": ["https://x.com/NASA/status/2107095768821576019"],
  "transcriptionMethod": "ai",
  "segmentGranularity": "word",
  "includeWordTimestamps": true
}

Example record

{
  "type": "video transcript",
  "tweetUrl": "https://x.com/NASA/status/2107095768821576019",
  "tweetId": "2107095768821576019",
  "authorUsername": "NASA",
  "authorName": "NASA",
  "language": "en",
  "transcriptionMethod": "ai",
  "transcriptAvailable": true,
  "modelUsed": "base",
  "segmentGranularity": "segment",
  "segmentCount": 1,
  "segments": [
    {
      "index": 0,
      "start": 0.0,
      "end": 11.0,
      "text": "One, booster ignition, and liftoff, the crew of Artemis II, now bound for the moon, humanity's next great voyage begins."
    }
  ],
  "fullTranscript": "One, booster ignition, and liftoff, the crew of Artemis II, now bound for the moon, humanity's next great voyage begins.",
  "publishedAt": "2026-10-05T13:09:13+00:00",
  "collectedAt": "2026-10-06T18:54:33+00:00"
}

Whisper models and memory

Model	Speed	Accuracy	Recommended run memory
tiny	Fastest	Lowest	1024 MB
base	Fast	Good (default)	1024 MB
small	Medium	Better	2048 MB
medium	Slow	High	4096 MB
large-v2	Slowest	Best	8192 MB

Long videos are split into 10-minute chunks automatically, so there is no video length limit. The actor runs with 8192 MB by default and never below 2048 MB. If the selected model does not fit the available memory, it automatically uses the next smaller model and records the substitution in modelNote instead of crashing. Caption-only runs have no meaningful memory requirement.

Use cases

• Content accessibility — produce subtitles (SRT/VTT) for public videos
• Media monitoring — transcribe interviews, press clips and announcements into searchable text
• Research — build text corpora from public video content for analysis
• Newsrooms — quote-check and index video statements quickly
• Podcast and clip workflows — repurpose spoken content into articles or show notes
• Social listening — combine transcripts with engagement data to track narratives

FAQ

Do I need an account or cookies? No. There is no cookie input; public posts, captions and media are read without login.

What happens when a video has no captions? With transcriptionMethod: auto (the default) the actor transcribes the audio with Whisper, so you still get a transcript. With captions it emits a row with transcriptAvailable: false and a message.

Can I transcribe multiple videos in one run? Yes — add links to postUrls, or point the actor at profiles and it transcribes the latest video post of each.

Which output format should I choose? segments for timestamped lines, fullText for one clean block of text, both for both, and srt/vtt when you need subtitle files (both the subtitle text and a signed download link are returned).

When does word-level granularity work? Word rows are emitted only when real per-word timings exist — Whisper can produce them, and some X captions include them. When they are missing, the row falls back to raw segments and explains why in granularityNote. Timings are never invented.

Can I transcribe videos in other languages? Yes — native captions work in any language X provides them; for AI, add a language hint (for example es, fr, ur) or leave it empty to detect automatically.

Does this support Twitter Spaces? No — this actor handles video and GIF posts only, not live audio Spaces.

Why is a row missing segments or fullTranscript? outputFormat controls which fields are filled, and unavailable transcripts never contain transcript fields — they are status rows with a message.

Can a deleted or protected post be transcribed? No, only videos on public, currently-live posts can be read.

Is this affiliated with X? No. This is an independent tool and is not affiliated with, endorsed by, or sponsored by X Corp.