Remove watermarks from videos via drag-and-drop. Uses LaMa inpainting. SVGOMG-style UX: drop a file, draw mask rectangles, download the result.
# simple-lama-inpainting pins Pillow <10; install it separately first
pip install simple-lama-inpainting --no-deps
pip install -e .
markdry serveThe browser opens at http://localhost:7777 automatically.
On first run, the LaMa model (~196 MB) is downloaded once to ~/.cache/torch/.
- Python 3.11+
ffmpegandffprobeon$PATH(brew install ffmpeg)- Apple Silicon: CoreML/MPS is selected automatically, CPU otherwise
- Drop a video onto the page (MP4, MOV, MKV — up to 500 MB).
- Drag rectangles over the watermark on the preview frame.
- Click Process. Progress streams frame-by-frame.
- Click Download clean video when done.
Audio is preserved. Processing uses the same static mask for every frame.