TrustMark

This repository contains the official, open source implementation of TrustMark watermarking for the Content Authenticity Initiative (CAI) as described in:

TrustMark - Universal Watermarking for Arbitrary Resolution Images (arXiv:2311.18297) by Tu Bui¹, Shruti Agarwal², and John Collomosse^{1 2}.

and later published at ICCV 2025 as TrustMark: Robust Watermarking and Watermark Removal for Arbitrary Resolution Images.

Overview

This repository contains the following directories:

• /python: Python implementation of TrustMark for encoding, decoding and removing image watermarks (using PyTorch). For information on configuring TrustMark in Python, see Configuring TrustMark.
• /js: Javascript implementation of TrustMark decoding of image watermarks (using ONNX). For more information, see TrustMark - JavaScript implementation.
• /rust: Rust implementation of TrustMark. for more information, see TrustMark — Rust implementation.
• /c2pa: Python example of how to indicate the presence of a TrustMark watermark in a C2PA manifest. For more information, see Using TrustMark with C2PA.

Model files (ckpt PyTorch file for Python and onnx ONNX file for JavaScript) are not packaged in this repository due to their size, but are downloaded upon first use. See the code for URLs and md5 hashes for a direct download link. Note that the location path was updated from Netlify to S3 in April 2026.

More information:

• For answers to common questions, see the FAQ.
• For information on configuring TrustMark in Python, see Configuring TrustMark.

Changelog

For a history of notable changes to the Python implementation, see the CHANGELOG.

Installation

Prerequisite

You must have Python 3.8.5 or higher to use the TrustMark Python implementation.

Installing from PyPI

The easiest way to install TrustMark is from the Python Package Index (PyPI) by entering this command:

pip install trustmark

Alternatively, after you've cloned the repository, you can install from the python directory:

cd trustmark/python
pip install .

Quickstart

To get started quickly, run the python/test-decode.py script that provides examples of watermarking several image files from the images directory.

Run the example

Run the example as follows:

cd trustmark/python
python test-decode.py

You'll see output like this:

Initializing TrustMark watermarking with ECC using [cpu]
Extracted secret: 1000000100001110000010010001011110010001011000100000100110110 (schema 1)
PSNR = 50.357909
No secret after removal

Example script

The python/test-decode.py script provides examples of watermarking a JPEG photo, a JPEG GenAI image, and an RGBA PNG image. The example uses TrustMark variant Q to encode the word mysecret in ASCII7 encoding into the image ufo_240.jpg which is then decoded, and then removed from the image.

from trustmark import TrustMark
from PIL import Image

# init
tm=TrustMark(verbose=True, model_type='Q') # or try P

# encoding example
cover = Image.open('images/ufo_240.jpg').convert('RGB')
tm.encode(cover, 'mysecret').save('ufo_240_Q.png')

# decoding example
cover = Image.open('images/ufo_240_Q.png').convert('RGB')
wm_secret, wm_present, wm_schema = tm.decode(cover)

if wm_present:
   print(f'Extracted secret: {wm_secret}')
else:
   print('No watermark decoded')

# removal example
stego = Image.open('images/ufo_240_Q.png').convert('RGB')
im_recover = tm.remove_watermark(stego)
im_recover.save('images/recovered.png')

Data schema

TrustMark encodes a payload (the watermark data embedded within the image) of 100 bits. You can configure an error correction level over the raw 100 bits of payload to maintain reliability under transformations or noise.

In payload encoding, the version bits comprise two reserved (unused) bits, and two bits encoding an integer value 0-3 that specifies the data schema as follows:

• 0: BCH_SUPER
• 1: BCH_5
• 2: BCH_4
• 3: BCH_3

For more details and information on configuring the encoding mode in Python, see Configuring TrustMark.

Citation

If you find this work useful, please give us a star ⭐ and cite the repository and/or TrustMark paper, preferrably as follows:

@inproceedings{Trustmark-ICCV-2025,
  title = {TrustMark: Robust Watermarking and Watermark Removal for Arbitrary Resolution Images},
  author={Bui, Tu and Agarwal, Shruti and Collomosse, John},
  booktitle = {IEEE International Conference on Computer Vision (ICCV)},
  year = {2025},
  month = oct
}

or cite the earlier ArXiv version:

@article{Trustmark-ArXiv-2023,
title={Trustmark: Universal Watermarking for Arbitrary Resolution Images},
author={Bui, Tu and Agarwal, Shruti and Collomosse, John},
journal = {ArXiv e-prints},
archivePrefix = "arXiv",
eprint = {2311.18297},
year = 2023,
month = nov
}

License

This package is is distributed under the terms of the MIT license. The MIT license covers both the code in this repo and the model files which are downloaded upon first use.