Remove–AI–Watermarks – CLI and library for removing AI watermarks from images (github.com)

A new tool has been developed to remove visible and invisible watermarks from AI-generated images, specifically targeting outputs from models such as Google Gemini, ChatGPT/DALL-E, Stable Diffusion, and Midjourney. This tool effectively strips various watermarking techniques, including visible logos and invisible patterns embedded by models like SynthID and StableSignature, alongside associated metadata. Utilizing a combination of reverse alpha blending for visible marks and diffusion-based regeneration for invisible ones, the tool offers both speed and batch processing capabilities, making it a versatile option for users seeking cleaner images. The significance of this development lies in its implications for the AI/ML community and content creators. By enabling the removal of watermarks that trigger "Made with AI" labels on social platforms, it raises ethical questions about originality and authorship in AI-generated content. Additionally, while the tool serves legitimate purposes such as privacy protection and art preservation, it could also foster misuse in deceptive contexts if users aim to obscure AI origins. The ongoing debates around watermarking laws and regulations further complicate this landscape, highlighting the need for responsible usage within the creative sectors and beyond.