<p>Driven by the development of artificial intelligence and computers, due to the needs of entertainment and social interaction, the application of artificial intelligence in synthesizing comic images from one's own photos is very extensive. At present, there are problems with using artificial intelligence to synthesize comics, such as a single style and poor exaggerated facial images. In response to the existing problems, this study proposes a fast facial image transfer algorithm based on multi-scale style transfer and semantic segmentation. This study involves transferring styles to each type of face and recombining and exaggerating facial features. This study proposes a facial exaggeration method based on statistical information, which formalizes the specific rules and calculation methods of facial feature point group exaggeration. A comparative analysis was conducted on the effectiveness of the image generated by the research algorithm and other algorithms. The experimental results showed that the proposed facial stylization and facial shape exaggeration methods have achieved good results. Compared with other algorithms, when the image size was 256 × 256 and 512 × 512, the computational efficiency was 7.62 s and 12.35 s, respectively, which was lower than the computation time of other algorithms. Research has shown that facial image conversion algorithms provide theoretical support for the future application of artificial intelligence in comic creation.</p>