Subtitles first
When a video already has subtitles, they're used directly; no speech recognition needed.
Sample lecture · Linear Algebra: Eigenvalues and Eigenvectors
The sample lecture was made for this page; its channel and video are fictional.
Free and open source · MIT
2.0.1What's new
Turn lecture videos into illustrated notes
Every lecture you've watched, ready to flip through again
course2md turns a whole lecture into an illustrated handout: a screenshot each time the slide changes, with what the lecturer said about it underneath. For YouTube and Bilibili videos, every section links back to that moment.
A screenshot of every slide, with what was said underneath.
Reviewing online courses
Self-study with open courseware
Talks and lab-meeting recordings
Apple Silicon · macOS 15 or later · also available via Homebrew
Intel Mac? Use the command-line version
“But look at v, which points along the diagonal. After the transformation it is still on the same dashed line.”
From “I watched it” to “I can use it”
Slides and explanation, together
Slides alone don't explain the derivation, and subtitles alone have no pictures. Here every screenshot sits right above what the lecturer said about it.
Formulas, typeset
With AI on, formulas, even spoken ones, are written as LaTeX and typeset. Copying a passage keeps each formula's LaTeX.
Bilingual notes
AI can add a translation under every paragraph, in any of eight languages.
Your AI, or none
Use any OpenAI-compatible service, a local Ollama server or your ChatGPT subscription. Off until you turn it on.
Queue up a whole course
Pick several parts of a multi-part Bilibili video or a playlist at once. Tasks run in the background; pause, resume or reorder them.
Your notes, your files
Plain Markdown with screenshots, in a library on your computer with nested folders and title search. Export a web page any time. No account needed.
Made to fit
Ten light and dark themes, adjustable text size, and English, 简体中文, 繁體中文 or 日本語.
More of the app
Your audio stays on your computer
Speech is recognized locally by default. Audio or text leaves your computer only if you choose a cloud speech service or an AI service.
Local recognitionDefault
Cloud speech serviceOptional
AI proofreading and summariesOff by default
Download the desktop app
Everything needed to convert videos is included. The first local transcription downloads a speech model (1–2.4 GB).
2.0course2md 2.0 is a rebuilt desktop app: preview before converting, background tasks you can pause and reorder, a library with folders, a reader with typeset formulas, and four interface languages. Release notes
macOS · Apple SiliconmacOS 15 or later
Download DMG
Or install with Homebrew and keep it current with brew upgrade
$
brew install --cask mizorewww/tap/course2md-gui
Windows · x64Portable package: extract everything to a permanent folder, then open course2md-desktop.exe
Download ZIP
GPU and CPU speech recognition need the Microsoft Visual C++ runtime; if it's missing, Run checks links Microsoft's installer.
Arch Linux / CachyOS · x86_64Installs ffmpeg, yt-dlp and the graphics libraries. Needs a working Vulkan driver; local speech recognition also needs the llama-cpp package.
$
yay -S course2md-gui-bin
Ubuntu and other distributions · x86_64Built on Ubuntu 24.04. Install the graphics libraries separately (see the Linux guide).
Download tar.gz
System requirements
- macOS
- Apple Silicon, macOS 15 or later
- Windows
- x64
- Linux
- x86_64 with a working Vulkan driver; the tar.gz is built on Ubuntu 24.04
- Tools
- Everything needed to convert videos is included. The AUR package uses Arch's ffmpeg, and local speech recognition there needs the optional llama-cpp package. On Windows, GPU and CPU recognition also need the Microsoft Visual C++ 2015–2022 runtime.
- Disk
- 1–2.4 GB for the speech model, downloaded once
- Intel Mac, Linux ARM64
- No desktop app yet; use the command-line version.
See the installation guide for details.
Prefer the terminal?
The command-line tool does the same conversion and installs alongside the desktop app. The two share your Bilibili and ChatGPT logins and the downloaded speech models; each keeps its own settings. Handy for batches and scripts, and it also runs on Intel Macs and Linux ARM64.
curl -fsSL https://raw.githubusercontent.com/mizorewww/course2md/main/install.sh | bash
CLI guide
~/courses ❯ course2md https://www.youtube.com/watch?v=…
Reading video information…
Looking for subtitles…
Downloading video…
Scanning video frames…
Saving screenshots…
Extracting audio…
Transcribing speech…
Writing notes…
Notes ready: out/youtube/3f9a…/versions/…/
Tip: run course2md llm setup to enable AI proofreading and summaries
~/courses ❯
Questions
Is course2md free?
Yes. It's open source under the MIT license, with no account, subscription or ads.
Do I need an account?
No. You only sign in to services you choose to use: a Bilibili account for some videos, or an AI service if you turn AI on.
Does the AI cost money?
AI is optional and off by default. With an OpenAI-compatible service you pay that provider for what you use; a local Ollama server costs nothing; a ChatGPT subscription works through the Codex login. course2md itself charges nothing.
Where are my notes and data?
In a library folder on your computer: each note is a course.md file with its screenshots, and you can export a web page, a .zip or JSON at any time. Add more library folders in Settings → Storage. Service keys are kept in the system keychain. Nothing is uploaded unless you choose a cloud speech or AI service.
How large are the speech models?
About 1–2.3 GB on Apple Silicon and 2.4 GB on other systems, downloaded once. In mainland China they come from domestic mirrors, and downloads resume after interruptions.
A Bilibili video fails or comes back in low quality
Log in under Settings → Services & accounts → Bilibili account by scanning the QR code.