ai-watermarking

KO | EN

gitlite — search

#hacktoberfest #typescript #machine-learning #docker #macos #java #cli #kubernetes #vue #pytorch #mcp #openai #ios #open-source

ai-watermarking

(★ 10)

AI watermarker for images and text that replicates SynthID's functionality.

ai-watermarking Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Recreating SynthID: Watermarking and Tokenizing AI-generated Content (Images and Text)

[![Watch how I made it](https://img.youtube.com/vi/WTQxY6Iw56A/0.jpg)](https://www.youtube.com/watch?v=WTQxY6Iw56A)

## Inspiration

AI has been accelerating the spread of misinformation and disinformation online, and it’s been harder to tell AI-generated and human-generated text apart. Roughly 57% of AI-generated content on the internet is most likely AI-generated. At the same time, tons of photorealistic AI-generated images have been fooling the internet despite algorithms designed to demote and remove AI-generated content.

Recently, Google unveiled SynthID: an algorithm that inserts digital watermarks in AI-generated content, including images, videos, text, and audio. These watermarks can be used to track AI-generated content circulating the internet. For my second replicate at The Knowledge Society, I recreated SynthID's basic function for watermarking images and text. Feel free to use this code in any AI-watermarking projects as long as you follow the Apache 2.0 license.

**DISCLAIMER: Please note that this project replicates the way that SynthID watermarks and detects AI-generated text on a smaller scale, and does not have the full functionality of SynthID.**

Read more on Medium about this project and how watermarking works: https://medium.com/@consigli/ai-vs-reality-whats-real-watermarking-1c1d2277d2db

## Currently this repository consists of code for:

* An image watermarking system (for images generated by Stable Diffusion)
* A system to determine the likelihood that tokens were generated by AI (specifically GPT2 by using tokenization)

## Requirements:

The Jupyter Notebooks in this repository can be used in Google Colab or locally. If you are interested in running this locally, make sure to install the latest version of [Python](https://www.python.org/downloads) that is [compatible with PyTorch](https://pytorch.org/get-started/locally/) and the necessary libraries:

```
pip install torch transformers pillow diffusers
```

## Acknowledgments

I wanted to give a special thanks to Sumanth Dathathri and Pushmeet Kohli who helped me better understand how SynthID works as I continue exploring it for my focus on Artificial Intelligence at The Knowledge Society!

ai-watermarking

ai-watermarking Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

ai-watermarking

ai-watermarking Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README