With the rapid development of technology and media in recent decades, the issue of art work generation has attracted significant attention from the academic community. To address the optimization of recursive algorithms in art work generation, a multi-scale generative adversarial network (PRMGAN) based on progressive recursion has been designed, along with the corresponding loss function. Building on this, the DualGAN network is employed to model artistic style transfer in art work generation. Finally, the model proposed in this paper is applied to conduct an in-depth analysis of art generation and artistic style transfer. Under the influence of PRMGAN and DualGAN, the metric values of the generated artworks have improved, but the effects are not significant, with corresponding SSIM and PSNR values ranging from [0.1 to 0.9] and [7 to 28], respectively. However, under the combined influence of the two, the metric values of the generated artworks have improved significantly, with SSIM and PSNR values ranging from [0.1 to 0.9] and [7 to 28], respectively. This study not only advances the field of art generation but also provides theoretical references for art style transfer modeling pathways.
