Show HN: Remove the Gemini watermark by solving the alpha blend, not inpainting (github.com)

A new tool has been launched that allows users to remove the Gemini watermark from images generated by Google’s Gemini model directly on their own machines, without uploads or external dependencies. This functionality is available as a browser tool, TypeScript library, and CLI, enabling users to recover the original image data obscured by the watermark. The method leverages the alpha blending equation to reverse the watermarking process, restoring the original pixels without any inpainting or guesswork involved. The tool can handle images in various formats and operates exclusively on locally stored files. This development holds significance for the AI/ML community as it provides a legitimate way for users to obtain clean canvases of their generated content while maintaining the integrity of invisible provenance watermarks like SynthID. The approach not only promises exact pixel recovery but also clearly establishes boundaries by ensuring that it doesn't attempt to defeat AI-content detection systems. The tool respects creators’ rights, focusing on content that users have personally generated rather than enabling broader manipulation of copyrighted material. The increasing interest in maintaining visual integrity in AI-generated content showcases the evolving landscape of user agency in artificial intelligence applications.