wiltodelta/remove-ai-watermarks

Remove visible and invisible AI watermarks and provenance metadata from images and video. Python library and CLI for SynthID, C2PA, EXIF, IPTC, XMP, and common generative-AI marks.

What it solves

This tool removes AI-generated provenance marks from images and videos that you own. It targets three types of markers: visible labels (like the Gemini sparkle), invisible pixel-level watermarks, and AI-specific metadata (C2PA, EXIF, XMP, etc.).

How it works

Depending on the target, the tool uses different techniques:

  • Visible Removal: Detects registered marks in expected areas, creates a mask, and fills the region using OpenCV, MI-GAN, or LaMa inpainting.
  • Metadata Removal: Uses format-aware stripping to remove AI tags without recompressing images or altering video stream offsets.
  • Invisible Removal: Employs diffusion pipelines (such as Qwen-Image-2512 or SDXL) to regenerate the image or video sequence, disrupting the pixel and frequency domain watermarks.

Who it’s for

Content creators who generate AI media and wish to remove provenance signals from their own files for a cleaner presentation.

Highlights

  • Comprehensive Video Support: Handles visible marks from Sora, Veo, Seedance, Doubao, Dola, Hailuo AI, and Kling AI.
  • Multi-Backend Inpainting: Supports OpenCV, MI-GAN, and LaMa for high-quality visible mark removal.
  • Invisible Watermark Disruption: Uses CUDA-accelerated diffusion regeneration to target hidden watermarks.
  • Metadata Stripping: Cleans C2PA, EXIF, XMP, IPTC, and China TC260 AIGC labels across various video containers (MP4, MOV, MKV, etc.).
  • Batch Processing: Includes CLI commands for processing entire directories of images or videos.

Related

  • Dispatch
    Breaking the Seal: The Technical Battle Against AI Watermarking

    An exploration of the 'remove-ai-watermarks' tool, analyzing the technical methods used to strip visible, invisible, and metadata-based AI provenance markers.

  • Project
    guillaumemeyer/watermarks-remover

    A service and agent skill for stripping AI provenance marks and watermarks from text and files across multiple vendors and formats.

  • Project
    allenk/GeminiWatermarkTool

    A cross‑platform, portable executable that removes Google Gemini image (and video) watermarks using a deterministic reverse‑alpha‑blending algorithm, with optional GPU‑accelerated FDnCNN denoising, GUI/CLI interfaces, batch processing, and AI‑agent integration via MCP.

  • Dispatch
    OpenAI Content Provenance Updates

    OpenAI is implementing a multi-layered provenance system using C2PA conformance, Google SynthID watermarking, and a new public verification tool to increase the transparency and durability of AI-generated content signals.

  • Dispatch
    AI Watermarking 101: Tools and Techniques

    Hugging Face provides a comprehensive overview of AI watermarking techniques across images, text, and audio to combat deepfakes and ensure content provenance.