With the increased use of light-emitting diode technology, the design freedom of taillight shapes has improved, and the shape design of taillights has become a major design focus and breakthrough point for new energy vehicles. As traditional taillight shape design normally depends on the individual experience of designers who make considerable effort to capture design inspiration by manually searching for design references, taillight shape creative design in the field of computer-aided design is studied to assist designers' creative thinking and design efficiency. In this research, generative adversarial networks are applied to explore the creative generation mechanism of taillight shape designs. First, a dataset of high-quality vehicle taillight shape images is established, which combines safety and aesthetic taillight shape design rules and insights. Then, deep convolutional generative adversarial networks and progressive growing of generative adversarial networks are both introduced to train and construct the taillight shape generative design model, and the commonalities and differences between the two are further compared. Finally, the optimal creative generation design model of the taillight shape is constructed. This research provides a new idea for taillight shape design, and the constructed taillight shape creative generation design model is helpful in realizing efficient, diversified and innovative taillight shape design and provides designers with rapid and sustainable preliminary design proposals and creative inspiration support.