Deep Convolutional Generative Adversarial Network (DCGAN) for Artificial Portrait Art Images

When AI generated art was first making waves, I decided to experiment with some of the early models that had come out. Before diffusion models became popular, generative adversarial networks were by far the most promising models for image generation. This project is my own attempt to develop and train a stable GAN for portrait art generation.

The model was trained on images from the Met Museum's collection of portrait art images. Additionally, the model's architecture and hyperparameters are the results of my own experimentation. Code and a sample gif from a walk through the image latent space can be found here.