This project is about generating album cover art using StyleGAN trained on a dataset of metal cover arts, and generating titles using an image captioning model.
Includes a simple Flask web application for generating samples and style mixing.
The training data (cover images and album titles) is automatically gathered from the publicly available databases of MusicBrainz and Cover Art Archive.
Title generation via image captioning is modeled by a pretrained ResNet as the encoder and a character-level LSTM as the decoder.
