TikTok Video Subtitles API

Endpoint facts

Method

GET

API path

/tiktok/mobile/video/subtitles

Auth

apiKey

Credits

8 credits/request

Charged on

Successful 2xx responses only

Platform

TikTok

Execution

Managed browser rendering where the target needs it

Response schema

#/definitions/tiktokmobile.videoSubtitlesResponseDoc

How this endpoint behaves

One video's own captions, translated into a language you choose, as TikTok's caption/language (cla) service serves them. The upstream is GET /tiktok/cla/subtitle_translation/get/v1/, captured from the running 47.4.3 app when the caption-translation control is used on a video and reproduced end-to-end against the live gateway.

Both a track and its cues. The response carries the captions twice: webvtt is the upstream WebVTT text verbatim, and cues is the same track parsed into {start_ms, end_ms, start, end, text} records, so a caller can build a transcript, burn subtitles into a render, or search caption text without implementing WebVTT. cue_count is the parsed cue count. Multi-line cues keep their line breaks in text.

Keyed by video id. video_id is the numeric aweme id that the feed, search, profile-posts and hashtag endpoints already return. You do not need TikTok's internal subtitle_id: the app sends one alongside the video id, but the server does not validate its value (measured: random values and even 1 returned the identical track), so this API sends the video id for it and hides it from callers.

Languages. language is required and validated upstream. Measured working codes (2026-10-21): ar, de, en, es, fr, hi, id, it, ja, ko, pt, ru, th, tr, vi, zh. Anything else is rejected locally with a 400 instead of paying a round trip for the upstream's status 5.

Device-free. The route is not signature-value gated: the pure-Go placeholder signer returned the same real tracks, so a deployment needs TIKTOK_MOBILE_PARAMS but not the device-backed signer (TIKTOK_MOBILE_SIGNER_URL). It answers on the core aggr16-normal.tiktokv.us / api16-normal-useast5.tiktokv.us pair.

Query parameters:

• video_id: required. The numeric aweme id of the video.
• language: required. Target language code from the measured set above.

Errors:

• 400 — the input is invalid, or the video carries no captions (the upstream answers status 5 "Invalid parameters" for a caption-less video).
• 500 — the upstream failed; a nonexistent video answers status 4 "Server is currently unavailable".

Example request:

GET /api/v1/tiktok/mobile/video/subtitles?video_id=7696141388324293918&language=en

The compact example shows two cues; the measured live track returned 11.

{
  "code": 200,
  "msg": "OK",
  "data": {
    "video_id": "7696141388324293918",
    "language": "en",
    "cue_count": 2,
    "cues": [
      {"start_ms": 80, "end_ms": 880, "start": "00:00:00.080", "end": "00:00:00.880", "text": "Hey."},
      {"start_ms": 3080, "end_ms": 4880, "start": "00:00:03.080", "end": "00:00:04.880", "text": "Our family is finally together again after so long."}
    ],
    "webvtt": "WEBVTT\n\n00:00:00.080 --> 00:00:00.880\nHey.\n\n00:00:03.080 --> 00:00:04.880\nOur family is finally together again after so long.",
    "variant": "glide_v3_1",
    "content_source": "redis",
    "host": "aggr16-normal.tiktokv.us",
    "fetched_at": "2026-10-21T10:56:17Z"
  }
}

Notes for callers:

• Track size is a snapshot, not a contract. The upstream serves a warm translation cache whose tracks can be re-translated, and the cue split varies by language (11–14 cues for this video). The same video's English track measured 750 chars / 9 cues at 19:53 and 766 chars / 11 cues at 22:30 on 2026-10-21; the language table above is the later measurement. Assert that the track is non-empty and in the requested language, not a byte count.
• The upstream serves a warm translation cache (content_source reported redis/db, translation_cache_time 7–13); measured latency ranged ~0.4–3.9 s per call across measurements.
• variant is the upstream's translation-pipeline tag and is omitted when the cache returns a track without one.
• If a language is requested for a video whose detected source language is that same language, the track can come back untranslated (the server's own fallback); the returned text is always the track the app would show.

Shared guides

Related