The aim of our project is to compare the ability of well-designed diffusion models to generate large synthetic financial correlation matrices with those obtained by DCGAN. We need to understand what they have managed to do and improve on it using new technologies. For example, image recognition, which has improved a great deal in recent years.