Comments and transcripts

Two commands cover the text around a video: comments reads the comment thread, transcript reads the captions. Both take a video id or URL, and both honor -n/--limit and --max-pages for paging, plus -o for the output format.

Comments

comments <video-id|url> streams the top-level comments for a video.

ytb comments dQw4w9WgXcQ

Flag	Meaning
--replies	Also fetch replies; each reply carries its parent_id
--all	Remove the default cap and page the whole thread
--sort	top or new

Newest 50 top-level comments:

ytb comments dQw4w9WgXcQ --sort new -n 50

Top comments with their replies threaded in:

ytb comments dQw4w9WgXcQ --replies

Comment bodies come from YouTube's current entity-payload model, so each row is parsed out of that representation rather than a flat list. With --replies, reply rows are tagged with the parent_id of the comment they answer, which lets you reassemble the thread downstream.

Restricted Mode

YouTube applies Restricted Mode to some server and datacenter IP ranges, and that mode hides comments entirely. When the response comes back gated, comments reports comments are hidden by Restricted Mode and exits with an error. It does not pretend the video has zero comments. On a normal residential connection comments stream normally. See Troubleshooting for the gating details.

Transcripts

Two commands, and the split is what a video has against what one track says. captions <video-id|url> lists the tracks; transcript <video-id|url> fetches one of them and writes it out.

ytb captions dQw4w9WgXcQ              # what tracks exist
ytb transcript dQw4w9WgXcQ            # joined text

Flag	Meaning
--format	text (default), srt, vtt, json
--lang	Caption language code
--auto	Take the auto-generated track
--translate	Ask YouTube to machine-translate the chosen track into this language code
--out	Write to this file instead of stdout

All four serializations come off one parse, so the timings in the srt and the json are the same timings.

ytb transcript dQw4w9WgXcQ --format srt --out never.srt
ytb transcript dQw4w9WgXcQ --format json | jq '.cues[0]'
ytb transcript dQw4w9WgXcQ | wc -w

Picking a track

The caption list is where the choice is made, and vss_id is the column that matters. .en is the track a person wrote and a.en is the one a machine wrote, which is what tells two tracks with the same language code apart.

With no --lang the human track wins, because a person's punctuation is worth more than a machine's word list. --auto overrides that and asks for the machine track even when a human one exists; it is the only track that carries per-word timings, which --format json keeps.

--lang matches loosely in the one direction that is safe: --lang es finds es-419 when there is no plain es.

ytb transcript dQw4w9WgXcQ --lang es

--translate asks YouTube for a machine translation of the chosen track into a language nobody has captioned it in. The translation is the site's own, which is worth knowing before quoting it.

ytb transcript dQw4w9WgXcQ --translate vi

No yt-dlp, no JavaScript

Transcripts used to need yt-dlp on your PATH, because YouTube gates the raw timedtext endpoint and a direct fetch came back empty. That is no longer true: the track is fetched and parsed in-process, with no yt-dlp, no Deno, and no JavaScript interpreter anywhere in the process tree.

env PATH=/usr/bin:/bin ytb transcript dQw4w9WgXcQ

A yt-dlp binary is still used as a fallback if the endpoint refuses and one happens to be installed, and it says so when it does. See Troubleshooting for the gating details.