Back to list
Comprehensive Open-Source Tool Remove-AI-Watermarks Enables Stripping of Visible and Invisible Identifiers from Major AI Models
Open SourceAI WatermarkingImage ProcessingGenerative AI

Comprehensive Open-Source Tool Remove-AI-Watermarks Enables Stripping of Visible and Invisible Identifiers from Major AI Models

A new open-source project titled "Remove-AI-Watermarks" has been released, providing a unified solution for removing both visible and invisible AI-generated markers from images. The tool targets outputs from industry-leading models including Google Gemini, OpenAI's DALL-E 3, Stable Diffusion, Adobe Firefly, and Midjourney. By utilizing techniques such as reverse alpha blending and diffusion-based regeneration, the library can eliminate SynthID, StableSignature, and TreeRing watermarks. Furthermore, it strips C2PA Content Credentials and EXIF/XMP metadata that trigger "Made with AI" labels on social media platforms like Instagram and Facebook. The project also introduces features like an "Analog Humanizer" and "Smart Face Protection" to maintain image quality while bypassing AI classifiers, offering both a command-line interface and a web-based service at raiw.cc.

Hacker News

Key Takeaways

  • Multi-Model Support: The tool effectively removes watermarks and metadata from Google Gemini (including Nano Banana and Gemini 3 Pro), ChatGPT/DALL-E 3, Stable Diffusion, Adobe Firefly, and Midjourney.
  • Visible and Invisible Removal: Employs reverse alpha blending for visible logos (like Gemini's sparkle) and diffusion-based regeneration for invisible watermarks such as SynthID and StableSignature.
  • Metadata Stripping: Targets C2PA provenance manifests, EXIF/XMP labels, and PNG text chunks to prevent automated "Made with AI" labeling on social media platforms.
  • Advanced Image Preservation: Features "Smart Face Protection" to prevent distortion during the AI cleaning process and an "Analog Humanizer" to bypass AI image classifiers.
  • Accessibility: Offers a batch processing library for local use and a free online web service via raiw.cc.

In-Depth Analysis

Technical Mechanisms for Watermark Erasure

The "Remove-AI-Watermarks" project introduces a multi-tiered approach to cleaning AI-generated imagery, categorized by the type of watermark encountered. For visible watermarks, such as the Google Gemini "sparkle" logo, the tool utilizes a fast, offline, and deterministic method known as reverse alpha blending. This technique effectively reverses the transparency-based overlay to restore the original pixel data beneath the logo.

For more sophisticated, invisible watermarks—including Google's SynthID (versions 1 and 2), StableSignature, and TreeRing—the tool employs diffusion-based regeneration. This process involves using AI diffusion models to subtly re-process the watermarked areas, effectively breaking the steganographic patterns or Discrete Wavelet Transform (DWT) markers that allow detectors to identify AI origin. The tool includes a three-stage Normalized Cross-Correlation (NCC) watermark detection system with confidence scoring to identify these hidden markers before removal.

Metadata and the Battle for Content Provenance

A significant portion of the tool's functionality is dedicated to stripping metadata that triggers automated labeling on social media platforms. Platforms such as Instagram, Facebook, and X (Twitter) increasingly rely on C2PA (Coalition for Content Provenance and Authenticity) manifests and specific XMP DigitalSourceType labels like "Made with AI" to inform users about the nature of the content.

Remove-AI-Watermarks targets a wide array of metadata formats, including EXIF, PNG text chunks, and C2PA provenance manifests across various file types such as PNG, JPEG, AVIF, HEIF, and JPEG-XL. By stripping these manifests and labels, the tool allows users to bypass the automated classification systems implemented by major tech companies. This includes the removal of prompt data, model versions, and seed information typically embedded in Midjourney or Stable Diffusion outputs.

Specialized Processing: Humanization and Face Protection

Beyond simple watermark removal, the library addresses the challenges of image quality and classifier detection. The "Analog Humanizer" feature adds film grain and chromatic aberration to images. This is specifically designed to bypass AI image classifiers that look for the characteristic smoothness or mathematical perfection of synthetic images, making them appear more like traditional analog photography.

To address the potential for AI-based regeneration to distort human features, the tool includes "Smart Face Protection." This feature automatically extracts human faces from the image and blends them back in after the watermark removal process. This ensures that while the background or metadata is cleaned of AI markers, the integrity and recognizability of human subjects are preserved, preventing the common "AI distortion" effect.

Industry Impact

The release of "Remove-AI-Watermarks" represents a significant development in the ongoing arms race between AI watermarking technologies and tools designed to circumvent them. As major AI providers like Google, OpenAI, and Adobe implement C2PA and SynthID to comply with safety standards and transparency goals, the availability of a "one-command" removal tool challenges the effectiveness of these industry-wide provenance standards.

This tool highlights the technical difficulty of maintaining permanent AI identifiers. By combining metadata stripping with pixel-level regeneration, it provides a blueprint for how AI-generated content can be decoupled from its origin. For the AI industry, this underscores the limitations of current watermarking techniques and may drive the development of more robust, tamper-resistant identification methods. Furthermore, it places additional pressure on social media platforms that rely on these markers for content moderation and user transparency.

Frequently Asked Questions

Question: Which AI models are currently supported by this tool?

The tool supports a wide range of models, including Google Gemini (Nano Banana and Gemini 3 Pro), OpenAI's DALL-E 3 and ChatGPT (including gpt-image-2), Stable Diffusion (via AUTOMATIC1111 and ComfyUI), Adobe Firefly, and Midjourney.

Question: How does the tool handle invisible watermarks like SynthID?

It uses a diffusion-based regeneration process to overwrite the imperceptible pixel patterns used by SynthID, StableSignature, and TreeRing. It also includes a detection system to identify these watermarks before the removal process begins.

Question: Can this tool prevent social media platforms from labeling my images as AI-generated?

Yes, the tool is specifically designed to strip C2PA manifests, EXIF/XMP "Made with AI" labels, and other metadata that trigger automated AI labeling on platforms like Instagram, Facebook, and X (Twitter).

Related News

OpenVINO on GitHub Trending: An Open Source Toolkit for AI Inference Optimization and Deployment
Open Source

OpenVINO on GitHub Trending: An Open Source Toolkit for AI Inference Optimization and Deployment

A GitHub Trending entry dated November 20, 2026, describes OpenVINO as an open source toolkit for optimizing and deploying AI inference. The entry credits openvinotoolkit and links to the OpenVINO repository on GitHub. Its description identifies two purposes for the toolkit: optimization and deployment. It does not explain how OpenVINO carries out either task or provide examples of its use. The supplied item also does not name a release, describe a new feature, or report benchmark results. The listing is therefore an introduction to the project and its stated purpose, not a release announcement or an assessment of how well the toolkit performs.

Univer Office SDK Brings Editing, Calculation, Collaboration, and File Processing to AI Office Products
Open Source

Univer Office SDK Brings Editing, Calculation, Collaboration, and File Processing to AI Office Products

A GitHub Trending entry from dream-num presents Univer Office SDK as a way to build an AI office inside a product. The description names four capabilities: editing, calculation, collaboration, and file processing. It positions those capabilities for both people and AI agents and says the SDK is built on an extensible open-source core. The central distinction is that Univer is presented as a toolkit for adding office functionality to a product, rather than as a finished office application. The brief entry does not describe how the capabilities are implemented, which file formats are supported, or how AI agents interact with the SDK.

Claude-BugHunter: A Comprehensive Red-Teaming Skill Bundle for Claude Code
Open Source

Claude-BugHunter: A Comprehensive Red-Teaming Skill Bundle for Claude Code

Claude-BugHunter, a trending GitHub project by elementalsouls, introduces a robust skill bundle tailored for Claude Code, specifically designed for bug hunting and external red-team operations. This toolset streamlines security assessments by integrating 82 distinct skills, 15 slash commands, and an extensive library of 681 disclosed-report patterns spanning 24 core vulnerability classes. Furthermore, the bundle enhances security testing by incorporating enterprise identity and infrastructure attack matrices, providing users with a structured framework for complex security analysis within the Claude Code ecosystem.