YouTube Transcripts & Subtitles — Text, Timestamps, SRT & VTT

Turn explicit public YouTube video URLs or IDs into real caption text and timed segments for research, accessibility workflows, search indexing and subtitle processing. Pick preferred caption languages, prefer human or generated tracks, and export text, segments, SRT or WebVTT from the same result row.

This Actor reads existing public caption tracks. It does not listen to audio, generate missing speech recognition, log in, retrieve private videos or bypass age/subscriber restrictions. It does not discover videos from channels, playlists or keyword search. Video title/channel/view enrichment and machine translation are outside this version.

First transcript

{"videos":["jNQXAC9IVRw"],"languages":["en"],"maxVideos":1}

The default input requests one short public video. The input schema's global delivery limit is3, but one supplied ID can only produce one useful row. YouTube access varies by IP and region; a video with captions in your browser is not guaranteed to be available from Apify.

Useful workflows

Goal	Result
Collect video text for research	Clean caption text in the selected real source language.
Link text to video moments	Start, duration and end per caption segment.
Export subtitle files	Optional SRT and VTT strings in the same paid row.
Prefer a language	Ordered languages with explicit optional fallback; actual selected language is always returned.
Process a known video list	Canonical video-ID deduplication, small parallel waves and a global transcript limit.

Input settings

Setting	Default and limit	Behavior
videos	One example;1–1000	Video ID, watch URL, youtu.be, Shorts, embed or public live-replay URL. Playlists/channels/search are rejected.
languages	["en"]; up to20	Actual caption track codes in preference order. No automatic translation.
preferManual	true	Human captions before generated within a matched language. False prefers generated.
fallbackToAnyLanguage	true	Use an available track when preferred codes are absent. Disable for strict language matching.
maxVideos	3; up to1000	Maximum delivered useful video transcripts. Failures do not consume result slots.
maxConcurrency	2;1–5	Small fetch waves; proxy traffic/compute for prefetched undelivered videos can still cost resources.
includeSegments	true	Add start/duration/end/text segments; plain text is always returned.
includeSrt / includeVtt	false	Add subtitle text formats; no separate result event for each format.
proxyConfiguration	Apify RESIDENTIAL	Residential proxy traffic is additional and user-paid. Direct access can be attempted with useApifyProxy=false but was unreliable in cloud QA.

Ready-to-copy scenarios

Known video batch

{"videos":["jNQXAC9IVRw","aircAruvnKk","Ks-_Mh1QhMc"],"maxVideos":3}

Subtitle text formats

{"videos":["https://youtu.be/jNQXAC9IVRw"],"maxVideos":1,"includeSrt":true,"includeVtt":true}

German first, then English, without another-language fallback

{"videos":["aircAruvnKk"],"languages":["de","en"],"fallbackToAnyLanguage":false,"maxVideos":1}

Compact plain text

{"videos":["Ks-_Mh1QhMc"],"includeSegments":false,"maxVideos":1}

Prefer an automatic caption track

{"videos":["jNQXAC9IVRw"],"languages":["en"],"preferManual":false,"maxVideos":1}

Human/automatic preference does not create a missing track: when only one matching type exists, that available track is returned and labelled. A successful English fallback is not advertised as a German transcript.

What each useful row contains

Field	Meaning
videoId, url	Stable requested video identity and canonical watch URL.
language, languageName	Actual returned caption track, not the requested first preference if fallback occurred.
isAutoGenerated	Source track classification. Automatic captions can contain recognition errors.
availableLanguages	Tracks visible to this request, with codes/names and generated flags.
text, segmentCount	Caption text and count of nonempty segments.
segments	Optional {start,duration,end,text}; seconds from the selected source captions.
srt, vtt	Optional subtitle strings of the same selected track.
source, scrapedAt, fetchAttempts	Caption provenance, observation time and1–2bounded fetch attempts.

No fabricated transcript is returned for a missing video. Caption timing is source timing, not word-level alignment or a promise of transcription accuracy. Omitting segments does not change the underlying selected track or create a cheaper second source.

Errors, partial runs and continuation

The default Dataset holds useful transcripts only. Failed-video IDs and codes live in ERRORS; SUMMARY describes delivered, failed and remaining videos. Missing tracks, strict-language misses, unavailable videos, source IP blocks and timeouts are distinct diagnostics. An inaccessible source is not an assertion that the video has no captions.

A run with no useful transcripts and a source/network/proxy failure fails visibly. An accessible no-caption/no-matching-language batch can finish with NO_TRANSCRIPTS and zero transcript rows. Partial runs retain useful output and explicit failures. Check diagnostics instead of trusting platform SUCCEEDED alone.

After an item or charge cap, submit CONTINUATION.resumeInput for pending IDs. A temporary source/proxy failure gets at most one in-run retry with a fresh sticky proxy session; no-caption and strict-language misses are not retried. After both attempts fail, review the diagnostic before any explicit new run. CHECKPOINT is written after processed items. When a run crashes during Dataset delivery or before its checkpoint save, first reconcile Dataset video IDs; do not blindly repeat uncertain items and bill them again. This version does not guarantee atomic delivery/checkpoint or automatically restart a crashed run.

Pricing and resource costs

Result tariff: Free $2.00 per 1,000 useful video transcripts; all paid tiers $1.50 per 1,000. Text, segments, SRT and VTT share one result event. Actor Start is $0.00005 per GB of memory, minimum one. Platform resources and optional proxy traffic are additional and paid by the user. Read the Pricing tab for effective account rates.

One useful transcript row is one result even with segments, SRT and VTT. Diagnostics, duplicates, unavailable/no-caption videos and undelivered transcripts do not create transcript result events. Compute and proxy traffic can still cost money on failed or empty runs. Download the completed Dataset in other formats rather than starting another collection.

Start with512MB and900seconds for a small batch. Each source request has a25second timeout. Large inputs, long captions, blocking and proxy latency can exceed run limits or memory;1000is an input/delivery ceiling, not a verified1000video performance promise. The source adapter documents cloud-IP restrictions; direct and proxy reliability must be measured separately. The default uses Apify residential proxy, which requires proxy access on your account and charges traffic separately. A three-video source probe returned3/3 through residential versus1/3 directly; this small sample is not a success-rate guarantee. No external subscription or customer account cookies are used.

API, MCP and automation

Exact slug: apivault_labs/youtube-transcripts-scraper. Use asynchronous Actor runs, wait for a terminal status, then read the Dataset plus SUMMARY/ERRORS/CONTINUATION. Keep your Apify token in secure credentials. Dataset export supports JSON/CSV/Excel; SRT/VTT are strings per useful video row.

Official Python client, JavaScript client and Apify MCP can pass this input to the exact Actor. They are generic clients, not dedicated packages for this new product. No matching dedicated npm/PyPI or verified n8n node is currently confirmed. HTTP workflows in n8n/Make are possible; execution/verification is not implied.

Five useful scenario Tasks are prepared separately for verification. Public one-click availability depends on actual publication; private Tasks are not public templates. Use the JSON scenarios above even when public Tasks are unavailable.

FAQ

Can it transcribe a video without subtitles? No. It reads available captions and never invents audio-derived text.

Can it translate captions? No. Language preference and fallback select existing tracks; actual selected language is explicit.

Does it support channels/playlists? This version accepts explicit video IDs/URLs only. A playlist query on a valid watch URL does not expand other videos.

Why did a video work locally but fail in Apify? YouTube can treat datacenter/proxy IPs differently. Inspect SOURCE_BLOCKED or other diagnostics. No success guarantee follows from local access.

Do manual captions always win? Language priority wins first. Within the first matching language, preferManual chooses manual if available. False prefers generated; missing types fall back to a matching available track.

Are live broadcasts supported? Only a public replay with available captions may work. An ongoing live stream without a complete caption track is not promised.