Introduction Diffusion models represent a cutting-edge advancement in the field of generative modeling， they offer a novel approach to learning complex data distributions by progressively corrupting training data with Gaussian noise and subsequently denoising it.

The Variational Autoencoder (VAE) is a probabilistic approach that extends the conventional autoencoder framework by introducing stochasticity in encoding and decoding processes, thus enabling the modeling of complex distributions of data.

Generative Adversarial Networks(GAN) has become one of the most powerful techniques in machine learning since its emergence in 2014. The interesting idea of training methods and the flexible design of the objective function make GAN have numerous variants.