AI-powered language learning from videos with Anki
CardBunny transforms video scenes into Anki flashcards automatically. It accepts YouTube links or local videos, transcribes dialogue using AI, translates it, and slices the media into digestible, reviewable study cards.
- 🎥 Video Slicing: Automatically generates WebM video clips for each spoken sentence.
- 🎵 Audio Extraction: Saves the exact scene audio in MP3 format.
- 🖼️ Scene Snapshots: Captures a high-quality JPG image of the scene.
- 📝 AI Transcription: Uses Faster-Whisper to accurately transcribe spoken dialogue.
- 🌐 AI Translation: Connects to any OpenAI-compatible API to translate sentences into your target language.
- ⚡ Seamless Integration: Directly pushes generated cards and media to Anki via AnkiConnect.
- Download: Fetches the video from YouTube or loads a local file.
- Transcribe: Uses Faster-Whisper to identify and transcribe all speech regions.
- Audit: Automatically audits and skips non-speech or silent regions.
- Translate: Pushes the dialogue through an OpenAI-compatible API endpoint.
- Slice: Uses FFmpeg to generate exact video clips, audio files, and images for every sentence.
- Sync: Sends the media and notes directly to your Anki deck through AnkiConnect.
- Windows 10 or 11
- Python 3.10+
- Anki running with the AnkiConnect add-on installed
- FFmpeg and FFprobe available in your
PATH - An OpenAI-compatible translation service (like Ollama, LMStudio, or OpenAI API)
Clone the repository and run the setup script:
setup.batThis script creates the venv virtual environment and installs all required dependencies.
To start the application, simply run:
run.batOr launch it directly using python:
venv\Scripts\python.exe src\main.py- Open Anki and ensure AnkiConnect is active.
- Start CardBunny.
- Paste a YouTube URL or select a Local Video.
- (Optional) Select a local
.srtsubtitle file to bypass transcription. - Choose your target Deck and Note Type.
- Map the generated content (Video, Audio, Source Text, Translation) to your Anki fields.
- Click the Play button to begin processing!
- YouTube URL
- Local video file
- YouTube URL + Local subtitles
- Local video + Local subtitles
Click the Gear icon in the app to configure your integrations:
| Section | Options |
|---|---|
| Anki | AnkiConnect address, default deck, and default note type. |
| Local AI | Endpoint, translation model, API key, Whisper model size, compute device, precision, and CPU threads. |
| Languages | Video source language and target translation language. |
Note: Settings are saved in
config.json. You can easily import/export settings directly from the interface.
For a balanced CPU configuration, we recommend:
{
"model_size": "medium",
"device": "cpu",
"compute_type": "int8",
"cpu_threads": 8
}The model is downloaded automatically on first use. Use small for faster processing, or large-v3 for maximum accuracy.
CardBunny creates a temporary workspace for each run:
workspace_<id>/
├── video.mp4
├── original.srt
├── translated.srt
├── transcription_audit.txt
└── slices/
├── *.webm
├── *.mp3
└── *.jpg
When processing finishes successfully, the application will prompt you to clean up or preserve these files.
- Cannot find Anki: Open Anki, ensure AnkiConnect is installed, and check that the default address is
http://127.0.0.1:8765. - Translation service not responding: Open settings and verify your Endpoint URL, Model Name, and API Key.
- First run is very slow: The Whisper model must be downloaded on its first run. It will be cached locally for all subsequent runs.
- Batch failed partway: CardBunny automatically halts the batch and attempts to roll back any notes/media sent during that run to prevent incomplete cards.
To generate a standalone .exe, ensure you have installed the PyInstaller dependencies inside the virtual environment and run:
venv\Scripts\pyinstaller.exe --noconfirm CardBunny.specThe resulting build will be located at dist\CardBunny\CardBunny.exe.
(Note: Always distribute the complete dist\CardBunny directory, as the executable depends on the _internal folder).