YouTube Video Transcript Scraper: Captions, Subtitles, SRT
Pricing
from $2.50 / 1,000 transcripts
YouTube transcript extractor for any video, channel, playlist or search: the video transcript text and timed segments, each track labelled manual or auto-generated captions. YouTube subtitles as SRT, WebVTT, Markdown or chunks, with metadata, chapters, comments, translation. No login.
YouTube videos in any form: watch, youtu.be, Shorts, live, embed and music.youtube.com links, or bare 11-character ids. One run takes many; repeats of the same video are read once. A channel or playlist link pasted here is read as one. A link to a text file, a CSV or a published Google Sheet with one URL or id per line is read as a list.
[]Channels to read the latest videos of: @handle, a channel link, or a UC… channel id. Newest first, up to maxVideosPerSource each.
[]Playlists: a playlist link or its id (PL…). In playlist order, up to maxVideosPerSource each.
[]YouTube searches: the videos the search returns, in YouTube's order, up to maxVideosPerSource each.
[]How many videos to take from each channel, playlist and search (videos without captions count; videos outside your dates or durations do not). Videos you list one by one are all read.
Which of a channel's uploads to read.
Only videos published on or after this date: 2026-09-01, or a span back from today such as 30 days, 2 weeks, 6 months. Applies to channels, playlists and searches; a channel's walk stops at the first video older than this.
Only videos published on or before this date (same forms as dateFrom). Applies to channels, playlists and searches.
Skip shorter videos from channels, playlists and searches. 0 = no limit.
Skip longer videos from channels, playlists and searches. 0 = no limit.
Transcript languages to look for, in order of preference: en, de, pt-BR, es-419. A code without a region also matches the regional track (de finds de-DE). When the video has none of them, the row carries the video's spoken language and language_note says so. Empty = the video's spoken language.
[]YouTube has captions written by people (manual) and its own speech recognition (auto-generated); every row says which it is (track_kind). The two "only" choices return a free labelled row when the video lacks that kind.
Also put every other caption track of the video on the row (other_tracks: each language, manual and auto-generated, with its text and segments). With trackKind manual only or auto-generated only, the other tracks are of that kind only. Charged per extra track.
Formats added to each row next to the plain text and the timed segments, which every row has.
[]About how long each RAG chunk is. Chunks end at caption boundaries and at chapter starts, never inside a caption.
Keep the timed segments (segments: start, duration, text) on each row. Off = text and formats only, for smaller rows.
Also write the text and each extra format as files in the run's key-value store (<videoId>.txt, .srt, .vtt, .md, …), with their links on the row (files). Free.
A language code (fr, de, ja, pt-BR, zh-Hans): each transcript is also translated by an OpenAI model and delivered next to the original (translation), segment for segment with the same timing, labelled machine (OpenAI gpt-6-luna). It is not YouTube's own translation, which YouTube does not serve logged out. Charged per started 1,000 characters of the original, only when every segment came back; a translation with a missing segment says which and costs nothing.
For a video without any caption track, transcribe its audio with OpenAI whisper-1 (timed segments), labelled speech-to-text (OpenAI whisper-1). Up to about 105 minutes of audio per video. Charged per started minute.
Top-level comments to add to each row (comments: text, author, likes, replies, when). 0 = none. Charged per comment.
The most transcripts this run delivers (and charges). 0 = no limit. Free rows (no captions, unavailable) do not count.
Skip videos this task delivered a transcript for in an earlier run: run a channel daily and get only its new videos. A video without captions is read again on each run (a free row): its captions can come later. Turns the memory on.