In the design process of traditional green and low-carbon building schemes, it is difficult to fully consider the impact of environmental factors, which limits the promotion and application of green and low-carbon buildings. This study uses Generative Adversarial Networks (GAN) technology to explore a new method that can automatically generate and optimize green and low-carbon building schemes. This paper designs a generator network structure, takes random noise as input, and gradually generates a candidate set of green and low-carbon building schemes through multi-layer convolution and transposed convolution operations. Then, a discriminator network structure is constructed, which takes the generated green and low-carbon building scheme as input, extracts the characteristics of the scheme through multi-layer convolution and pooling operations, and evaluates its authenticity and environmental protection. During the training process, an adversarial training strategy is adopted, and the generator and the discriminator are constantly confronted and iterated. Finally, the generated green and low-carbon building schemes are evaluated and verified. Compared with traditional design schemes, the carbon emissions of the GAN-generated schemes are lower, the energy saving rate is improved, and the sunshine compliance rate reaches 99.9%. In addition, through multi-objective optimization verification and industry expert evaluation, the superiority of the GAN-generated scheme in terms of innovation and practicality is further proved.