# Changelog of TikTok Transcript Scraper | Captions + AI Fallback (`zrk1212/tiktok-transcript-scraper`) Actor

- **URL**: https://apify.com/zrk1212/tiktok-transcript-scraper/changelog.md
- **Full Actor documentation**: https://apify.com/zrk1212/tiktok-transcript-scraper.md

## Changelog

### 1.0 (2026-11-17)

- First version: transcripts from TikTok video URLs, profiles (yt-dlp listing) and other scrapers' datasets. Native captions first (claInfo-based track choice: original language, creator vs auto captions, translations only on request), Groq Whisper fallback for paying users only.
- AI money controls: pay-per-event + price required on the platform, locked minute reservations against the spending limit, 2,000 AI minutes per run cap, ffmpeg cut at the reserved minutes, idempotent per video across migrations, circuit breaker on short charges, no AI charge for NO_SPEECH.
- Datacenter proxy: TikTok WAF challenges about 8 to 33% of datacenter IPs (1.5 KB page, HTTP 200). Up to 3 fresh datacenter sessions, then 2 residential.
- Measured locally: caption coverage 37/55 (67%) on 11 big accounts; Groq 0.4 s for a 29 s clip.
- Platform measurements (build 1.0.2, 1024 MB, 2026-11-17), 30 videos from 10 accounts, captions only, 100% page success in every mode:
  - no proxy: 61 s, $0.0020 (about $0.07 per 1,000 videos)
  - datacenter (default): 96 s, 34 page fetches (4 WAF retries), $0.0040 (about $0.13 per 1,000)
  - user-selected RESIDENTIAL: 111 s, 66 page fetches, 4.75 MB residential, $0.0289 (about $0.96 per 1,000)
  - 3x run, 90 videos datacenter: 134 s, $0.0068 (about $0.08 per 1,000), native share 65/85, 5 photo posts as free errors.
- Deviation from plan line 115 (approved): NO_SPEECH rows (AI found only music or silence) are charged the video-metadata price, not AI minutes.
- A user-selected RESIDENTIAL proxy group is replaced by datacenter-first (residential stays the automatic fallback), measured at about $0.96 per 1,000 videos when residential is used for every page.
- AI on the platform is gated until pay-per-event is configured, so AI cost is estimated: Groq about $0.0007 per minute, media over datacenter (no per-GB fee), AI skipped when the page needed residential. Verify with one paying platform run after PPE is set.
