Why the scope is narrow
Generic watermark removers often promise every logo, text overlay, and video mark. This project instead models documented Gemini and Veo watermark sizes, anchors, alpha behavior, rendering variants, and frame behavior.
That narrower scope enables safer candidate validation and clearer failure behavior. The supported sizes and formats guide publishes the catalog the detector actually uses, and visible Gemini watermark vs SynthID sets out what a clean corner does and does not prove.
How quality is evaluated
The repository maintains image and video fixtures, crop sheets, frame samples, residual metrics, selection traces, benchmark scripts, and release gates. New patterns are treated as data-discovery problems before thresholds are changed.
The same principle shapes the product surface: when evidence is weak, the tool preserves your file instead of forcing an edit. Removing a Gemini watermark without losing quality compares that approach with cropping, blurring, and generic inpainting.
Open ecosystem
The engine is available as a web tool, Chrome extension, userscript, CLI, and npm SDK. The code, tests, and release history are public for inspection.
- Browser eraser for images and the video watermark remover for clips.
- Chrome extension and Tampermonkey userscript for Gemini itself.
- CLI and npm SDK for local automation.
- Guides for the evidence behind each claim.
Where we draw the line
The tool removes documented visible marks only. It does not remove SynthID, Content Credentials, or arbitrary third-party watermarks, and it does not make AI-generated media human-made. Responsible use covers the disclosure and rights questions that survive any pixel edit.
Last updated 2026-07-31