Generative adversarial networks (GANs)¹ are deep learning (DL) methods, which are in turn a type of machine learning. In recent years, DL methods have been applied extensively in medicine and in ophthalmology, mainly for image classification, for example, for detecting glaucoma,^2–5 age-related macular degeneration (AMD),^{2 6–9} diabetic retinopathy^{2 10–13} and retinopathy of prematurity.¹⁴ As the name suggests, GANs are used not to classify images but to generate images, and have two main components. The first ‘generative’ network uses the training data to generate synthetic images, which are then presented to the second ‘discriminative’ network that is responsible for discriminating between the synthetic and real images. The two networks are ‘adversarial’ in that the ‘generative’ network aims to generate synthetic images that can ‘fool’ the ‘discriminative’ network. These two networks are then trained reiteratively against each other to ultimately maximise the ‘authenticity’ of the synthetic images. GANs have been applied in ophthalmology in several contexts. For example, they have been used to generate colour fundus photographs with different stages of AMD,¹⁵ improve the segmentation of anterior segment optical coherence tomography (OCT),¹⁶ create autofluorescence images …