What it does

AI-Media2Doc is a Vue-based web app that ingests video or audio and spits out structured documents in Chinese social-media styles — think Xiaohongshu posts, WeChat articles, knowledge notes, mind maps, or plain subtitles. It runs entirely locally via Docker, requires no user registration, and keeps task history in the browser’s local storage. The author built it because he prefers reading text to watching videos and didn’t want to hand his data to yet another SaaS platform.

The interesting bit

The “smart screenshot” feature is a neat hack: instead of calling a vision model (expensive, slow), it uses subtitle timestamps to grab frames from the video and inserts them inline at zero extra cost. FFmpeg runs in the browser via WASM, so you don’t even need it installed locally. The whole thing is pitched as “extremely low cost” — a deliberate contrast to the usual login-and-credit-card AI tools.

Key highlights

• Self-hostable with Docker; MIT licensed.
• No accounts needed; task records stay in local storage.
• Custom prompts configurable from the frontend.
• Subtitle export and AI Q&A against the transcribed content.
• Optional access-password lock for the backend.
• Roadmap includes fast-whisper local model support to cut cloud transcription costs further.

Caveats

• The README is entirely in Chinese; English speakers will need translation help for setup.
• Backend configuration requires Volcano Engine (火山引擎) environment variables; it’s unclear how vendor-locked that makes the LLM and transcription backends.
• Windows users are explicitly told to use WSL, so native Windows Docker support may be rough.

Verdict

Worth a look if you’re a Chinese-speaking content creator, researcher, or compulsive note-taker who wants to own their pipeline. Skip it if you need polished multilingual support or a managed SaaS with a support hotline.