By applying computer algorithms, architectural generative design can achieve efficient and precise architectural planning and design. With the development of artificial intelligence technology, architects and industry experts are actively exploring deep learning algorithms, particularly neural network learning, in order to optimize the architectural design workflow and alleviate the workload of designers. Through thorough comparison and research, the author has discovered that Convolutional Neural Networks (CNN), in particular, have the widest range of applications. They can extract various features of buildings and classify these features, thereby assisting us in evaluating and optimizing design systems. By training on existing image data or 3D model data, Generative Adversarial Networks (GANs) can effectively learn from the datasets provided by people. Through the adversarial interplay between the generator and discriminator, GANs continually refine their learning accuracy and subsequently generate new image data.