Luma Subtitle is a desktop app for generating, translating, and exporting video subtitles. After you import a video, the app extracts audio with FFmpeg, transcribes it locally with whisper.cpp, translates the subtitles through an OpenAI-compatible /v1/chat/completions API, and exports standard SRT files.
It is built for individual creators, course editors, interview workflows, and multilingual content production. Transcription, subtitle translation, model setup, dependency checks, task queues, and exports all live in one local desktop workspace.
The macOS build targets Apple Silicon. Automatic FFmpeg and whisper.cpp builds require macOS 11.0 or later, Xcode Command Line Tools, and cmake; whisper.cpp is built with the Metal backend.
- Video import and task management: choose video files, output folders, source language, target language, and translation settings.
- Local transcription: run whisper.cpp on your machine and select local Whisper model files.
- Subtitle translation: use any OpenAI-compatible Chat Completions API with configurable Base URL, model name, and API key.
- SRT export: generate source and translated subtitle files for editors, players, and subtitle tooling.
- Task queue: batch transcribe, translate, and export; optionally enable automatic chaining from transcription to translation to export.
- Environment panel: check FFmpeg, whisper.cpp, model folders, dependency folders, and download supported presets.
- Platform focus: Windows x64 and macOS Apple Silicon.
- Video processing, audio extraction, and whisper.cpp transcription run locally.
- Translation sends subtitle text to the OpenAI-compatible API endpoint configured by the user.
- API keys are stored in the local SQLite database under the app user data directory.
- Do not commit local models, FFmpeg/whisper binaries, task artifacts, development logs, personal settings, or API keys.
- Tauri 2
- React 18
- TypeScript
- Vite
- Rust
- whisper.cpp
- FFmpeg
- Windows x64: the app selects a CUDA whisper.cpp package when an NVIDIA GPU is available, otherwise it uses a BLAS/CPU package.
- macOS Apple Silicon: the app first uses installed or bundled arm64
ffmpegandwhisper-cli; if missing, it can build FFmpeg and Metal-enabled whisper.cpp from official source archives.
Intel Mac is not currently supported.
- Node.js 20+
- pnpm 9+
- Rust 1.80+
- Windows: NVIDIA driver and CUDA-capable GPU optional for faster transcription
- macOS Apple Silicon: macOS 11.0+, Xcode Command Line Tools, and
cmake
pnpm installThe in-app Environment panel shows the fixed dependency folder and model folder.
Clicking "Download to dependency folder" will:
- Windows: download and extract
ffmpeg.exeand the best matching CUDA, BLAS, or CPUwhisper-cli.exe. - macOS Apple Silicon: use installed or bundled dependencies first; if missing, download official source archives and build locally. It does not call Homebrew or download unofficial macOS binaries.
macOS automatic build requirements:
- Apple Silicon device running macOS 11.0 or later.
- Xcode Command Line Tools installed, with
clang,make,tar, andshavailable. cmakeinstalled for configuring and building whisper.cpp.- Network access to FFmpeg official source archives and
ggml-org/whisper.cppGitHub release source archives.
macOS build sources and configuration:
- FFmpeg is downloaded from
https://ffmpeg.org/releases/and built locally withVideoToolbox,AudioToolbox, andAVFoundationenabled. - whisper.cpp is downloaded from official
ggml-org/whisper.cppGitHub release source archives and built with CMake usingGGML_METAL=ON. - whisper.cpp explicitly uses the
macOS 11.0deployment target for Apple Silicon and C++17std::filesystemcompatibility.
To skip in-app compilation, pre-bundle dependencies in the release process or install them to the user's PATH. During development, you can also place executables at:
src-tauri/resources/bin/macos-arm64/ffmpegsrc-tauri/resources/bin/macos-arm64/whisper-cli
Release builds bundle files under src-tauri/resources. macOS executables must keep executable permissions:
chmod +x src-tauri/resources/bin/macos-arm64/ffmpeg
chmod +x src-tauri/resources/bin/macos-arm64/whisper-cliLookup order: app data directory, bundled resources, common macOS executable paths, then system PATH.
Whisper models can live anywhere. Select the model file in the app. On Apple Silicon, large-v3-turbo-q5_0 or small are good starting points depending on memory and speed requirements.
The in-app model presets download into the app data directory's models folder and automatically update the selected Whisper model path. You can also download these files manually and select them in the app:
| Preset | File | Size | Download |
|---|---|---|---|
| tiny | ggml-tiny.bin |
75 MiB | https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-tiny.bin |
| base | ggml-base.bin |
142 MiB | https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.bin |
| small | ggml-small.bin |
466 MiB | https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-small.bin |
| large-v3-turbo-q5_0 | ggml-large-v3-turbo-q5_0.bin |
547 MiB | https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-large-v3-turbo-q5_0.bin |
pnpm tauri:devThe same command works on macOS.
pnpm tauri:buildFor production macOS distribution, handle .icns icons, codesigning, and notarization on a macOS machine.
Luma Subtitle uses the official Tauri updater plugin. Release builds publish signed update artifacts and latest.json to GitHub Releases.
Generate the updater signing key once:
pnpm tauri signer generate -w ~/.tauri/luma-subtitle.keyStore the private key content in the GitHub secret TAURI_SIGNING_PRIVATE_KEY. If you protect the key with a password, store it in TAURI_SIGNING_PRIVATE_KEY_PASSWORD. The public key is committed in src-tauri/tauri.conf.json.
The app checks:
https://github.com/csic21/luma-subtitle/releases/latest/download/latest.json
Each task can generate:
{video_name}.source.srt{video_name}.{target_language}.srt
Intermediate task files are written under .luma-subtitle-work inside the output folder.
Before committing, make sure you do not include:
.env,.env.*, private keys, certificates, tokens, or real API keys.node_modules/,dist/, orsrc-tauri/target/.- Local development logs, layout-check screenshots, Whisper models, or FFmpeg/whisper binaries.
Luma Subtitle is licensed under the GNU General Public License v3.0 or later. See LICENSE for details.