Image style transfer refers to converting an image from one style to another while keeping the semantics of the content unchanged. Deep neural networks have excellent feature extraction and representation capabilities. It is widely used in the field of image style conversion. In this paper, the existing deep image style transfer models based on neural network are classified by analyzing the algorithmic principles of different image style transfer models. Then, the same test images are selected to test and evaluate the effect of different models. The evaluation results show that because of the powerful global modeling feature, the Transformer-based Str2 model has the best performance both in terms of computational efficiency and image style transfer quality. In the future, the Str2 model will be further improved in terms of feature interaction module as well as loss construction, which is more meaningful to enhance the image style transfer effect.