At present, there is still room for evolution in style transfer of open source programs. This research uses open source code for style transfer on GitHub. In addition, it supports the development of online AI Attraction Page, Windows versions, Andorid platform, and Intel NCS. It also strengthens calculation and supports bases of multiple platforms. It is able to implement static style transfer on film, and speed up style transfer inferencing performance on web page. In addition, the literature review explores aesthetic perception elements and applies them to calculate parameter setting. The results of this study discover when the content image weight is 7.5 and the style image weight is 120, the inferenced image can retain characteristics of the original image, and come out with new blending style. Besides, to freeze the content and style image weight ratio, and increase the style image weight value to more than 10,000, the thin film color effect may appear. When there are 32 filters, the extracted color and style can show the most appropriate proportion and state. When the style size is adjusted to 410 × 256 and the content image is close in size, the original style features become more prominent. Finally, keep the style image free space at appropriately 25%, higher texture effect may occur after training.