ComfyUI Extension: Blind_watermark_Comfyui

Authored by TianDongL

Created

Updated

8 stars

A powerful blind watermark plugin for ComfyUI that allows you to embed and extract invisible watermarks in images using frequency domain techniques (DWT-DCT-SVD).

Custom Nodes (0)

    README

    ComfyUI Blind Watermark

    English | δΈ­ζ–‡ζ–‡ζ‘£

    A powerful blind watermark plugin for ComfyUI that allows you to embed and extract invisible watermarks in images using frequency domain techniques (DWT-DCT-SVD).

    This project is based on https://github.com/guofei9987/blind_watermark.git Production, thank you for the big shot's open source

    Features

    • πŸ”’ Invisible Watermarks: Embed watermarks that are imperceptible to the human eye
    • πŸ“ Text Watermarks: Embed and extract text messages
    • πŸ–ΌοΈ Image Watermarks: Embed and extract image watermarks
    • πŸ“² QR Code Support: Generate and decode QR codes for advanced watermarking
    • πŸ›‘οΈ Robust: Resistant to common image manipulations (rotation, cropping, scaling, compression)
    • βš™οΈ Advanced Controls: Fine-tune embedding strength, DCT coefficients, and processing modes
    • 🎨 ComfyUI Integration: Seamless workflow integration with native node support

    Installation

    1. Install the Plugin

    Navigate to your ComfyUI custom nodes directory and clone this repository:

    cd ComfyUI/custom_nodes
git clone https://github.com/TianDongL/Blind_watermark_Comfyui.git

    2. Install Dependencies

    Install the required Python packages:

    cd Comfyui_blind_watermark
pip install -r requirements.txt

    Required packages:

    • blind-watermark - Core watermarking library
    • qrcode - QR code generation
    • pillow - Image processing
    • pyzbar (optional) - QR code decoding

    3. Restart ComfyUI

    Restart ComfyUI to load the new nodes.

    Available Nodes

    Watermark Nodes

    1. πŸ“ Embed Text Watermark

      • Embeds invisible text into images
      • Configurable embedding strength (d1, d2)
      • Supports multiple DCT coefficient modes

    2. πŸ“– Extract Text Watermark

      • Extracts embedded text from watermarked images
      • Requires matching password and length parameters
      • Provides detailed extraction info

    3. πŸ–ΌοΈ Embed Image Watermark

      • Embeds an invisible image watermark into another image
      • Supports PNG, JPEG, and other formats
      • Configurable alpha blending

    4. πŸ” Extract Image Watermark

      • Extracts embedded image watermarks
      • Requires watermark dimensions (width Γ— height)
      • Outputs extracted watermark as image tensor

    QR Code Nodes

    1. πŸ“² Generate QR Code

      • Creates QR codes from text/URLs
      • Adjustable size and error correction levels
      • Outputs as image tensor for further processing

    2. πŸ” Decode QR Code

      • Decodes QR codes from images
      • Automatic detection and extraction
      • Returns decoded text/URL

    Quick Start Guide

    Example 1: Text Watermark Workflow

    Embed Text:

    1. Load your image using "Load Image" node
    2. Add "πŸ“ Embed Text Watermark" node
    3. Connect the image and enter:
      • Watermark Text: "Copyright 2025"
      • Password: 12345
      • d1/d2: Use defaults (36/20) or adjust
    4. Save the watermarked image

    Extract Text:

    1. Load the watermarked image
    2. Add "πŸ“– Extract Text Watermark" node
    3. Enter the same password and text length
    4. View extracted text in output

    Example 2: Image Watermark Workflow

    Embed Image:

    1. Load the original image
    2. Load your watermark/logo image
    3. Add "πŸ–ΌοΈ Embed Image Watermark" node
    4. Connect both images
    5. Set password and embedding strength
    6. Save the result

    Extract Image:

    1. Load the watermarked image
    2. Add "πŸ” Extract Image Watermark" node
    3. Enter password and watermark dimensions
    4. View extracted watermark

    Example 3: QR Code Watermark

    1. Generate a QR code with "πŸ“² Generate QR Code"
    2. Use it as watermark with "πŸ–ΌοΈ Embed Image Watermark"
    3. Extract and decode with "πŸ” Extract Image Watermark" β†’ "πŸ” Decode QR Code"

    Parameters Explained

    Password Parameters

    • password_img: Password for image encryption (integer)
    • password_wm: Password for watermark encryption (integer)
    • Both passwords must match between embed and extract operations

    Embedding Strength (d1/d2)

    • d1: Controls frequency coefficient range (default: 36)
    • d2: Controls embedding strength (default: 20)
    • Higher values = stronger watermark, more visible
    • Lower values = weaker watermark, more invisible
    • Recommended: d1=36, d2=20 for balanced results

    DCT Coefficient Modes

    • Mode 1: Uses fewer coefficients (faster, less robust)
    • Mode 2: Uses more coefficients (slower, more robust)
    • Mode 3: Uses even more coefficients (slowest, most robust)

    Processing Modes

    • Single Process: Standard processing
    • Multi-core: Parallel processing for faster embedding
    • All Cores: Uses all available CPU cores

    Watermark Robustness

    This watermark technique is resistant to:

    | Attack Type | Robustness | Notes | |------------|-----------|--------| | πŸ”„ Rotation | ⭐⭐⭐⭐ | Survives moderate rotation | | βœ‚οΈ Cropping | ⭐⭐⭐⭐⭐ | Highly resistant | | πŸ“ Scaling | ⭐⭐⭐⭐⭐ | Survives resize operations | | 🎭 Masking | ⭐⭐⭐⭐ | Partial watermark still recoverable | | πŸ“¦ Compression | ⭐⭐⭐⭐ | Survives JPEG compression | | 🌈 Brightness/Contrast | ⭐⭐⭐⭐⭐ | Unaffected by color adjustments | | πŸ”Š Noise | ⭐⭐⭐ | Moderate resistance |

    Tips & Best Practices

    For Best Results:

    1. Password Management

      • Use the same passwords for embedding and extraction
      • Record passwords securely
      • Higher passwords = more security

    2. Watermark Length

      • Keep text watermarks concise (< 100 characters)
      • Longer watermarks require larger images
      • Note the reported wm_bit_length after embedding

    3. Image Quality

      • Use high-resolution images (minimum 512Γ—512)
      • Avoid heavily compressed images as input
      • Save watermarked images in lossless formats (PNG)

    4. Embedding Strength

      • Start with defaults (d1=36, d2=20)
      • Increase d2 if extraction fails
      • Decrease d2 if watermark is visible

    5. Image Watermarks

      • Use simple, high-contrast watermark images
      • Recommended watermark size: 64Γ—64 or 128Γ—128
      • Black and white logos work best

    Troubleshooting

    Watermark Extraction Failed

    Problem: Extracted text is garbled or empty

    Solutions:

    • βœ… Verify passwords match exactly
    • βœ… Check wm_bit_length parameter
    • βœ… Ensure image hasn't been re-encoded multiple times
    • βœ… Try increasing embedding strength (d2)
    • βœ… Use the same DCT mode for embed/extract

    Watermark Visible in Image

    Problem: Watermark creates visible artifacts

    Solutions:

    • βœ… Reduce d2 parameter (e.g., from 20 to 15)
    • βœ… Use a smaller watermark
    • βœ… Ensure input image is high quality

    Node Not Appearing

    Problem: Nodes don't show up in ComfyUI

    Solutions:

    • βœ… Check requirements.txt packages are installed
    • βœ… Restart ComfyUI completely
    • βœ… Check console for error messages
    • βœ… Verify folder is in ComfyUI/custom_nodes/

    Technical Details

    This plugin uses frequency domain watermarking based on:

    • DWT (Discrete Wavelet Transform)
    • DCT (Discrete Cosine Transform)
    • SVD (Singular Value Decomposition)

    The watermark is embedded in the frequency domain, making it invisible to the human eye while being detectable with the correct password.

    Credits

    Based on the excellent blind_watermark library by @guofei9987.

    Related Projects

    License

    This project follows the original blind_watermark library license. Please refer to the LICENSE file for details.

    Support & Contribution

    • πŸ› Bug Reports: Open an issue on GitHub
    • πŸ’‘ Feature Requests: Open an issue with your suggestion
    • πŸ”§ Pull Requests: Contributions are welcome!

    Happy Watermarking! πŸŽ¨πŸ”’