IndependentCompleted
vextract · Media Extraction Tool
Paste a share link, get video, audio, image sets or transcripts; custom providers with yt-dlp fallback
- Role
- Design and implementation (solo)
- Scale
- 7 custom providers + 1,700-site fallback
- Period
- Sep 2026 — Present
PythonFastAPIcustom providersyt-dlplocal transcription
What I did
- Hand-written providers for Douyin, Kuaishou, Xiaohongshu, Bilibili, Weibo, YouTube and TikTok, with yt-dlp covering the rest
- Zero re-encoding for audio: stream-copy when the source already matches the target format, tags included
- Fully offline transcription that needs no platform session
- Everything happens in the browser: paste → parse → pick quality → download → play in page
Why
Content work needs source material, and the available tools are paid, Windows-only, or video-only when all you want is the audio. So I wrote one.
Trade-offs
Measured, not claimed. What each platform supports — including how it fails — lives in a local verification record. An upstream README’s claim is not evidence.
Only what is needed. No session dependencies, no cloud, no accounts. A local server, a browser, done.
No gratuitous transcoding. An m4a source is copied as m4a; mp3 only when asked, with title and author tagged in.
Details
- An occupied port yields to the next free one, printing the process holding it
- The browser opens only after the server is listening and a health probe confirms it is this program
- Transcription runs entirely locally; audio never leaves the machine