- category
- watermark-remover
- maturity
- growing
- approach
- Open-source desktop tool that uses Microsoft's Florence-2 vision model to detect and localize watermark regions in an image or video frame, then applies LaMA (Large Mask Inpainting) to reconstruct the pixels underneath, rather than regenerating the whole image through a diffusion pass.
- access
- open-source
- owns
- A Python application with a PyQt6/PyWebview GUI and CLI for removing visible watermarks from images and videos, explicitly marketed for stripping visible marks left by generative video tools such as Sora and Runway, plus batch processing and audio-preserving video handling.
- distinctFrom
- Unlike remove-ai-watermarks or noai-watermark (which lean on diffusion-model regeneration to disrupt watermarks, including invisible pixel-level ones like SynthID) or wmr-gemini-watermark-remover (which inverts alpha-blend compositing mathematically), WatermarkRemover-AI's core technique is object-detection-plus-inpainting: it locates the visible watermark region with Florence-2 and reconstructs only that region with LaMA, and does not claim to defeat invisible or frequency-domain watermarks.
- notes
- MIT licensed; star counts vary by source snapshot (roughly 1,000 to 1,800 stars across secondary write-ups), so treat as 'over 1k stars' rather than a precise pinned number. Repository has an established commit history (50+ commits) and is covered by third-party roundups (scriptbyai.com, xugj520.cn) as of August 2026, but an exact launch date could not be confirmed; flagged as an untracked gap-fill rather than a confirmed new-in-window launch. Not previously tracked in this catalogue.
