Zero-shot learning (ZSL) is an AI model that can generate images that have not been labeled and presented in the training phase. At execution (inference), the model is given auxiliary descriptions.