Abstract: Recently,image generation and style transfer based on deep learning have been widely applied and there are lots of breakthroughs.In order to conduct research upon multi-style texture recovery of ancient books,we proposed a new structure of layout analysis and style fusion system in this paper.Firstly,we trained our models by using generative adversarial networks(GANs)and multi-style ancient books’background model to synthesize multi-style ancient textures;then,we analyzed layouts based on the position rearrangement(PR)algorithm to adjust the layout structure of foreground texts;finally,we realized the goal by fusing foreground texts and generated backgrounds.In the experiment,we chose ancient materials such as Yi scripts,ancient Chinese(seal),Jurchen scripts and ancient drawings as samples and improved the generation performance of different fine-turning model by improving DCGANs model in parameters as well as structures.Then,we evaluated the results using cross entropy loss function and Fréchet inception distance(FID).Eventually,we got model M8 with lowest FID.Compared with DCGANs model,the capability of M8 improved by 19.26%,enhancing the quality of the generated images profoundly.