The Idempotent Generative Network (IGN), jointly proposed by UC Berkeley and Google, is a new type of generative AI model that can generate realistic images in a single step. Unlike traditional Generative Adversarial Networks (GANs) and diffusion models, IGN is a self-adversarial model that accomplishes generation and discrimination in one step, mapping input to the target data distribution. Although the current results produced by IGN do not yet reach the level of state-of-the-art models, it is more efficient in inference, particularly in fields such as medical image restoration, where it has potential applications. Researchers have utilized a simple autoencoder architecture.