Open source

Built on open source.

OpaCaptions stands on open-source software and openly licensed AI models. This page lists them and gives you the source code the licences ask us to share. Last updated: 3 November 2026.

FFmpeg

The app runs FFmpeg as a separate program to read your files and write the captioned video. It is the unmodified LGPL build made by the FFmpeg-Builds project, licensed under the GNU Lesser General Public License version 3.

• Version: n8.1.3-25-g611cec792a (FFmpeg 8.1.3), win64 lgpl-shared, from the FFmpeg-Builds release autobuild-2026-10-28-19-47.
• FFmpeg source code for this version (tar.gz, 17 MB; SHA-256 d6228fe2111721062178221edf9b4c449b66fc9b06c1fd15eb14c97537daa559). Also on GitHub.
• The build scripts used to make it (tar.gz, 0.1 MB; SHA-256 31941af5905762d96b95760e44c7619a62a82a3f6df1d6c0c0ae67315efb3536). They name every library in the build and the exact source each one comes from.

You may replace the FFmpeg files in the app's ffmpeg folder with a build of your own. If you'd like a copy of the source of any library in the build, write to support@opacaptions.com and we'll send it.

AI models

• Speech recognition: OpenAI Whisper (MIT), in the CTranslate2 format by SYSTRAN, run by faster-whisper (MIT) with the Silero VAD speech detector (MIT).
• Captions in other languages: Meta's M2M100 418M (MIT).
• Speaker detection: NVIDIA TitaNet-S (Apache 2.0) through sherpa-onnx (Apache 2.0), and pyannote segmentation 3.0 (MIT).
• Magic Clips framing: the Ultra-Light-Fast-Generic-Face-Detector-1MB face detector (MIT).

Everything else

The app also uses Python, pywebview, Microsoft Edge WebView2, ONNX Runtime, CTranslate2, NumPy and about thirty other libraries, plus the caption fonts (SIL Open Font License and Apache 2.0). The full list with every licence text is in the app: Help > Third-party licences.

Thank you to everyone who builds and shares these.