Generate an NFT with Secret Messages

Summary

This project is a proof of concept for a NFT that can be used to store secret messages within a minted image that is stored on the blockchain as an NFT. The images are generated using a DCGAN model that was trained on a dataset of the famous CryptoPunk NFT's and the messages are encrypted using an AES-256 encryption algorithm. The NFT contract inherits from the ERC721 contract standard.

Some technologies that will be used are:

• TensorFlow/PyTorch ( GAN Model )
• Streamlit (Front end webapp)
• IPFS Protocol for NFT image storage.
• Steganography to encrypt secret data in the image.
• Ethereum Blockchain Based (For now)

Project Goals

• Generate 8-bit images based on the user input image.
• Provide an easy way to mint it to the blockchain. ( using front end webapp )
• Encrypting a user message within the image or a simple hashed version of the message.

Gif of GAN model generating images based on training data set

Model trained on RTX3070Ti, on full 10k image data set for 500 epochs using Tensorflow

Example of style transfer to GAN generated image

Instructions for running the code on your locacl machine

• Install the required dependencies using the command below pip install -r requirements.txt
• Install Ganache for local testing
• Using any online solidity compiler, compile the contract and deploy it to the local Ganache network.
• The contract abi is already provided in the main directory of the project folder.
• A SAMPLE.env is included in the project folder to access IPFS from Pianta
  • you must paste in your API key and the IPFS endpoint.
  • ensure to copy the deployed contract address from your preferred IDE used to compile the contract.
• Run the code using the command below streamlit run punkApp.py
• Follow the Demo video to see the app in action and learn how to interact with the app.

App Demo !

Contributors

• John Gaffney
• Jihad Al Hussain