GANpunks are 100% computer-generated creations, made using a Generative Adversarial Network (GAN) running in a homegrown Python environment. Creating the GANpunks images is a fully-automated Machine Learning process, with zero human intervention.

What you see in these collectible NFTs is the real-time process of a computer teaching itself how to draw a CryptoPunk.

To create each GANpunk, a Neural Network (NN) algorithm first generates a random set of 24x24 pixels. These random pixels are compared against the original CryptoPunk image, and a comparison error (CE) is calculated.

The NN then learns how to make itself smarter, and creates subsequent sets of pixels which result in lower and lower CE values. This process continues until the CE is so low that the computer-generated image matches the original CryptoPunk image.

This project is inspired by the CryptoPunks NFT collection created by Larva Labs.