I think they only mentioned the horse to illustrate to people that they are using the same tool as what was used to generate those types of images. It's painting a picture for the uninitiated audience. From what I understood, this model is trained on spectrograms instead of horses and the like, resulting in this product.
