With the rapid development of deep learning technology, GANs have made significant progress in the fields of image generation and style transfer. This paper proposes an improved GAN-based image generation and style transfer model, which combines the advantages of deep convolutional neural network (DCNN) structure and GAN, aiming to improve image generation and optimize style transfer quality. We verify the effectiveness of the model through comparative experiments and make detailed comparisons with traditional methods such as GAN, DCGAN and CycleGAN. Experimental results on multiple standard datasets (including CelebA, LSUN and artistic style transfer datasets) show that the proposed model performs well in terms of generated image quality, style transfer effect and training stability. Specifically, the proposed model has a FID (Frechet Inception Distance) score of 35.4, which is significantly lower than other comparison models (GAN: 55.8, DCGAN: 48.5, CycleGAN: 42.2), and also outperforms traditional models in SSIM (Structural Similarity Index) score (0.85 vs. GAN: 0.75, DCGAN: 0.78, CycleGAN: 0.80). These results show that the proposed model can generate higher quality and more stable images, and can better maintain the balance between content and style in style transfer tasks. Despite the study’s rather noteworthy findings, issues with model stability, style complexity, and training efficiency remain. Future studies will concentrate on resolving these issues in order to enhance the model’s scalability and application potential.