Video To Text, Subtitle Extraction, And Voiceover Script Rewriting

What Problem It Solves

When short videos are scattered across Bilibili, Douyin, WeChat Channels, and local folders, the hard part is turning them into reusable material: line-by-line transcripts, timed subtitles, and publishable voiceover drafts. media2text organizes one video into text, subtitles, and script drafts that downstream workflows can consume directly, reducing manual copying, screenshots, and re-locating source files.

How It Works

It uses a local ASR pipeline: detect whether the input is a URL or local file, select the download or parsing path, then generate results with faster-whisper. The base output is a UTF-8 .txt file with a metadata header and sentence-level line breaks. It can also produce .srt subtitles, and when a translation backend is configured, bilingual subtitles are available.
Common platform paths include:
- Bilibili: single videos and batch uploads from a creator channel, useful for extracting continuous spoken content.
- Douyin: video posts via cookie-free share links, image posts via extracted images and captions; audioless posts are not forced through transcription.
- WeChat Channels: local files can be processed directly; sph share links rely on a Yuanbao login session and produce time-limited direct links.
- Other and local files: general links are handled through yt-dlp, while local audio/video files go directly to local conversion and recognition.
Voiceover rewriting is enabled by default: for spoken-word videos, it outputs cleaned drafts that preserve key lines and add a hook. With DASHSCOPE_API_KEY, it can generate Bilibili, Douyin, and WeChat Channels tones; without a key, it falls back to a local cleaned draft. Rich mode creates a per-video folder and can output title, description, cover, no-watermark video, and manifest.json, making downstream TTS or editing tools easier to wire up.

Boundaries And Notes

It handles video-to-text and script drafting, not TTS, digital humans, or final rendering. Transcription quality depends on audio quality and model size; the default model is lightweight, while larger models can improve accuracy. Translation requires an external API key and degrades to source-only subtitles when missing. Private or age-restricted content may still need cookies. WeChat Channels sph links expire, so local files are the most stable fallback. Use should respect platform terms and content rights.