An offline AI eraser for baked-in subtitles and watermarks
It locally inpaints over hard-coded subtitles and text watermarks in videos and images so you never have to upload frames to a cloud API.

What it does
VSR is a local desktop application that detects and removes hard-coded subtitles and text-like watermarks from videos and images. Rather than blurring or tiling over the affected area, it uses AI inpainting models to synthesize replacement pixels while keeping the original resolution intact. It ships with both a GUI and a CLI, and can extract the original subtitle text for use with a companion extractor tool.
The interesting bit
The project treats text removal as a video restoration problem instead of a cheap cover-up. It bundles several distinct inpainting algorithms—STTN for live-action, LAMA for images and animation, and ProPainter for high-motion scenes—so you can trade speed against VRAM use and fidelity depending on the source. Everything stays on your machine, which matters if you are working with footage you would rather not upload.
Key highlights
- Runs fully offline with no third-party API keys or cloud uploads.
- Supports multiple AI inpainting backends: STTN, LAMA, ProPainter, and OpenCV.
- Can target specific subtitle regions by coordinates or attempt automatic full-text removal across the frame.
- Offers pre-built Windows packages for CPU, DirectML, and several CUDA versions, plus Docker images for Linux.
- Handles batch watermark removal on still images in addition to video.
Caveats
- On Apple Silicon Macs, the default PP-OCRv4-Server detection model is described as less accurate; the README suggests trying other models.
- STTN’s fastest mode can skip subtitle detection entirely, which risks missing text or accidentally erasing non-subtitle elements.
- ProPainter mode is explicitly noted to be slow and extremely VRAM-hungry despite its strength on high-motion video.
Verdict
Video editors and archivists who need to scrub burned-in text without re-encoding lossy cloud uploads will find this genuinely useful. If your needs are limited to cropping or blurring a static lower-third, traditional ffmpeg filters are still the lighter path.
Frequently asked
- What is YaoFANGUK/video-subtitle-remover?
- It locally inpaints over hard-coded subtitles and text watermarks in videos and images so you never have to upload frames to a cloud API.
- Is video-subtitle-remover open source?
- Yes — YaoFANGUK/video-subtitle-remover is open source, released under the Apache-2.0 license.
- What language is video-subtitle-remover written in?
- YaoFANGUK/video-subtitle-remover is primarily written in Python.
- How popular is video-subtitle-remover?
- YaoFANGUK/video-subtitle-remover has 12k stars on GitHub and is currently accelerating.
- Where can I find video-subtitle-remover?
- YaoFANGUK/video-subtitle-remover is on GitHub at https://github.com/YaoFANGUK/video-subtitle-remover.