Anki Gen

Generate Anki cards from anime episodes or other subtitled videos.

Requirements

• Python 3.9+
• ffmpeg on PATH

Install Python deps:

pip install -r requirements.txt

Usage

python -m anki_gen \
  --video "path/to/video.mkv" \
  --jp-srt "path/to/jp.srt" \
  --en-srt "path/to/en.srt"

Show full grouped help (with color in interactive terminals):

python -m anki_gen --help

You can force or disable ANSI coloring via --color always / --color never. If --format is omitted, output defaults to an Anki package (.apkg).

LLM-assisted review (two-step):

python -m anki_gen \
  --video "path/to/video.mkv" \
  --jp-srt "path/to/jp.srt" \
  --en-srt "path/to/en.srt" \
  --review-out \
  --user-level beginner \
  --llm-model gpt-5-mini

Set --llm-api-key (or env) to enable the LLM step. Set OPENAI_MODEL to change the default model globally. The CLI also auto-loads a .env file from the current working directory. Use --llm-provider to switch API providers (openai, openrouter, anthropic).

Example .env:

OPENAI_API_KEY=your_key_here
OPENAI_MODEL=gpt-5-mini
OPENAI_API_BASE=https://api.openai.com/v1
OPENAI_API_ENDPOINT=/chat/completions
OPENAI_TEMPERATURE=1.0
OPENAI_REASONING_EFFORT=minimal
OPENAI_DEBUG_FILE=output/llm_debug.log
LLM_PROVIDER=openai
LLM_MODEL=gpt-5-mini
LLM_API_KEY=
LLM_API_BASE=
LLM_API_ENDPOINT=
LLM_TEMPERATURE=1.0
LLM_REASONING_EFFORT=minimal
LLM_DEBUG_FILE=output/llm_debug.log
OPENROUTER_API_KEY=
OPENROUTER_APP_NAME=anki-gen
OPENROUTER_SITE_URL=https://example.com
OPENROUTER_API_BASE=https://openrouter.ai/api/v1
OPENROUTER_API_ENDPOINT=/chat/completions
ANTHROPIC_API_KEY=
ANTHROPIC_API_BASE=https://api.anthropic.com/v1
ANTHROPIC_API_ENDPOINT=/messages
ANTHROPIC_VERSION=2023-06-01

Provider examples:

# OpenAI (default)
python -m anki_gen ... --llm-provider openai --llm-model gpt-5-mini

# OpenRouter (OpenAI-compatible endpoint)
python -m anki_gen ... --llm-provider openrouter --llm-model openai/gpt-5-mini

# Anthropic
python -m anki_gen ... --llm-provider anthropic --llm-model claude-3-5-sonnet-latest

To debug LLM output formatting issues, add --llm-debug (stderr) and/or --llm-debug-file output/llm_debug.log.

The command above prints the generated review file path. Edit that TSV, flip the approved column as needed, then generate cards:

python -m anki_gen \
  --video "path/to/video.mkv" \
  --review-in output/20260214_220100_video-name/llm_reviews/review.tsv

With --review-in, card text uses reviewed fields:

• JP from focus (fallback: original jp)
• EN from gloss (fallback: reason, then original en)

For Anki packages:

python -m anki_gen \
  --video "path/to/video.mkv" \
  --jp-srt "path/to/jp.srt" \
  --en-srt "path/to/en.srt" \
  --out-dir output \
  --format apkg \
  --deck-name "Frieren S01E08"

Notes

• If --out-dir is omitted, output defaults to output/<timestamp>_<video>/ with:
  • llm_reviews/review.tsv (when using --review-out)
  • anki_output/deck.apkg or anki_output/cards.tsv
  • anki_output/media/audio and anki_output/media/img
• If --out-dir is provided, cards/media are written directly under that directory.
• The CSV export is a tab-separated file (cards.tsv).