Video-subtitle-remover is a tool for removing hard-coded subtitles and text watermarks from videos and images using AI-based inpainting.
The tool addresses the problem of permanently embedded subtitles in video files by using deep learning models to detect and remove text regions while reconstructing the underlying content. It performs inpainting on the areas where subtitles or watermarks are removed, filling them with plausible pixel values rather than simple interpolation or mosaic removal. The approach runs entirely locally without requiring third-party API calls. Users can either specify custom subtitle positions for targeted removal or enable automatic detection to remove all text from an entire video. The tool also supports batch processing of multiple images for watermark removal and can extract subtitles from videos for use with complementary tools.
The tool suits developers and users who need to process videos with hard-coded subtitles while preserving original resolution. It works across Windows, macOS, and Linux, with prebuilt packages available for different hardware configurations including CPU-only systems, non-NVIDIA GPUs via DirectML, and NVIDIA GPUs with various CUDA versions. Docker support is also provided. This approach is preferable to online services when privacy is a concern or when processing large volumes of video content locally is more practical than uploading to cloud services.
The project maintains multiple prebuilt distributions tailored to different computational environments and GPU capabilities, indicating attention to accessibility across hardware configurations. Development includes both a graphical interface and command-line interface for different usage patterns. The tool provides Docker containerization alongside traditional installation methods, suggesting consideration for deployment flexibility. Documentation is available in both simplified Chinese and English, reflecting effort to serve multiple language communities.
