- A new open-source AI watermark removal tool targets SynthID, C2PA manifests, and EXIF metadata across six major AI image platforms.
- AI watermark removal now covers both visible sparkle overlays and invisible frequency-domain patterns that survive cropping and JPEG compression.
- The tool strips the metadata that triggers ‘Made with AI’ labels on Instagram, Facebook, and X, raising serious content authenticity questions.
- An optional Analog Humanizer injects film grain and chromatic aberration to fool AI image classifiers entirely.
One Tool to Undo What the AI Industry Built
AI watermark removal just got a lot more serious. A GitHub project called remove-ai-watermarks has surfaced on Hacker News, and it’s not a half-baked script — it’s a full CLI tool and Python library capable of stripping both visible and invisible watermarks from images generated by Google Gemini, OpenAI’s DALL-E, Stable Diffusion, Adobe Firefly, and Midjourney. It handles the metadata layers that trigger “Made with AI” labels on social platforms, too. The implications for the entire AI content authentication ecosystem are worth taking seriously.
The project targets six distinct watermarking and provenance systems in one pipeline: Google’s SynthID, StableSignature, TreeRing invisible watermarks, C2PA Content Credentials, EXIF and XMP metadata fields, and the visible sparkle logo that Gemini overlays on generated images. That’s not a narrow exploit — that’s a near-complete attack surface against the current state of AI image identification.
How AI Watermark Removal Actually Works Here
The technical approach splits into two distinct tracks depending on whether the watermark is visible or invisible, and the engineering on both sides is genuinely interesting.
For visible watermarks — currently only used by Google Gemini, which the project notes is internally codenamed “Nano Banana” — the tool uses reverse alpha blending. Gemini composites its sparkle logo onto generated images using a standard alpha channel formula: watermarked = α × logo + (1 − α) × original. Since the alpha map is deterministic and extractable by running Gemini on a pure-black background, the math to reverse it is straightforward: original = (watermarked − α × logo) / (1 − α). A three-stage Normalized Cross-Correlation detector locates the watermark position and scale dynamically, so it works even if the image has been resized or cropped. The whole process takes roughly 0.05 seconds per image, requires no GPU, and runs entirely offline.
Invisible watermarks are a harder problem. Systems like Source: https://github.com/wiltodelta/remove-ai-watermarks
