Offline AI subtitle generation for macOS Apple Silicon.

Transcribes video audio using faster-whisper and burns subtitles using a bundled static ffmpeg with libass.

• 🔒 Fully offline — no internet required after first model download
• 🍎 Apple Silicon native — arm64 binary, optimized for M1/M2/M3
• 🎬 Subtitle burn-in — permanent subtitles via ffmpeg + libass
• 🎨 Customizable — font, size, color, outline, position
• 📦 Self-contained — no Homebrew, no system dependencies

• macOS 12.0+ (Monterey or later)
• Apple Silicon (M1/M2/M3)
• Python 3.10+ (arm64) — from python.org

Download from https://evermeet.cx/ffmpeg/:

• ffmpeg → place at assets/bin/ffmpeg
• ffprobe → place at assets/bin/ffprobe

chmod +x assets/bin/ffmpeg assets/bin/ffprobe

Verify:

# Must be arm64
file assets/bin/ffmpeg

# Must show ONLY /usr/lib/* — NO /opt/homebrew
otool -L assets/bin/ffmpeg

# Must show 'ass' filter
assets/bin/ffmpeg -filters 2>&1 | grep ass

Place TTF/OTF fonts in assets/fonts/:

cd assets/fonts
curl -fsSL -o NotoSans-Regular.ttf \
    "https://github.com/notofonts/noto-fonts/raw/main/hinted/ttf/NotoSans/NotoSans-Regular.ttf"
curl -fsSL -o NotoSans-Bold.ttf \
    "https://github.com/notofonts/noto-fonts/raw/main/hinted/ttf/NotoSans/NotoSans-Bold.ttf"

Place assets/icons/app_icon.icns for a custom icon. See assets/icons/README.md for creation instructions.

pip install -r requirements.txt

chmod +x build_app.sh
./build_app.sh

Output: dist/AutoSubtitle.app

chmod +x build_dmg.sh
./build_dmg.sh

Output: dist/AutoSubtitle-1.0.0-arm64.dmg

# Build .app
pyinstaller auto_subtitle.spec --clean --noconfirm

# Create DMG
codesign --force --deep --sign - dist/AutoSubtitle.app
mkdir -p dist/dmg_staging
cp -R dist/AutoSubtitle.app dist/dmg_staging/
ln -s /Applications dist/dmg_staging/Applications
hdiutil create \
    -volname "Auto Subtitle" \
    -srcfolder dist/dmg_staging \
    -ov -format UDZO \
    dist/AutoSubtitle-1.0.0-arm64.dmg
rm -rf dist/dmg_staging

dist/AutoSubtitle.app/
└── Contents/
    ├── Info.plist
    ├── MacOS/
    │   ├── AutoSubtitle              ← main executable (arm64)
    │   ├── bin/
    │   │   ├── ffmpeg                ← static arm64 ffmpeg (libass included)
    │   │   └── ffprobe               ← static arm64 ffprobe
    │   ├── assets/
    │   │   └── fonts/
    │   │       ├── NotoSans-Regular.ttf
    │   │       └── NotoSans-Bold.ttf
    │   ├── faster_whisper/
    │   │   └── assets/
    │   │       └── silero_vad_v6.onnx
    │   ├── PySide6/                  ← Qt frameworks
    │   ├── ctranslate2/              ← CTranslate2 inference engine
    │   └── *.so / *.dylib            ← Python extensions
    ├── Resources/
    │   └── app_icon.icns
    └── Frameworks/

Whisper models are downloaded to ~/.cache/huggingface/hub/ on first use. This requires internet on first run per model size.

After the model is cached, the app works fully offline.

Model sizes:

1. Open AutoSubtitle-1.0.0-arm64.dmg
2. Drag AutoSubtitle.app to Applications
3. Right-click → Open (first launch only, to bypass Gatekeeper)
4. Select a video file and click Start

xattr -dr com.apple.quarantine /Applications/AutoSubtitle.app

Ensure assets/bin/ffmpeg exists and is executable before building.

Verify ffmpeg has libass: assets/bin/ffmpeg -filters 2>&1 | grep ass

Check ~/Library/Logs/AutoSubtitle/app.log for details.