video-subtitle-extractor

Video Subtitle Extractor is a desktop tool that reads a video file and pulls out any text that has been burned directly into the picture, then saves those words as a standard subtitle file (SRT format) that media players can load separately. This is useful when you have a video with captions baked into the image rather than stored as a separate track, and you want to edit, translate, or search through that text.

The tool runs a text-detection process entirely on your own computer, so you do not need to sign up for or pay any online service. It supports 87 languages, including Chinese, English, Japanese, Korean, Arabic, French, German, Russian, Spanish, Portuguese, Italian, and more. You can process a single video or a whole folder of videos at once, as long as the videos share the same resolution and subtitle position.

There are three processing modes. Fast mode uses a lighter detection model and finishes quickly, though it may occasionally miss a line or misread a word. Auto mode picks the right model based on whether your computer has a compatible graphics card: lightweight when running on the CPU alone, more thorough when GPU acceleration is available. Precise mode checks every frame with the most accurate model but runs very slowly and is not recommended for routine use.

For graphics card support, the tool works with NVIDIA cards via CUDA, AMD and Intel cards via DirectML on Windows, and Apple Silicon or AMD GPUs via ONNX on macOS and Linux. A CPU-only mode is also available for machines with no compatible GPU at all.

If a video contains recurring on-screen labels, channel logos, or watermarks you want stripped from the output, you can list those strings in a small configuration file and the tool will remove or replace them in the extracted text. The project is open source under the Apache 2 license and provides both a graphical interface and a command-line version.