Integration and Application of Deep Learning Algorithms in Intelligent Image Visual Communication Design

Traditional image generation methods in visual communication design face challenges of inefficiency, inconsistent styles, and inadequate handling of complex requirements. To address these issues, this study proposes a novel framework integrating StyleGAN with cross-modal data fusion and interactive optimization. First, NLP converts ad copy into design instructions, guiding StyleGAN to generate prototypes. User behavior data and a curated visual element library enhance personalization and diversity. Transfer learning accelerates training while maintaining style consistency, supplemented by an attention mechanism to refine regional style control. An interactive module enables real-time feedback for iterative design adjustments. Experiments comparing StyleGAN and DALL•E demonstrate that StyleGAN achieves superior style consistency (SSIM: 0.86 vs. 0.74) and theme focus (36% vs. 24% element prominence), despite DALL•E’s faster generation. The results validate StyleGAN’s efficacy in high-quality, style-sensitive design tasks, offering practical insights for intelligent visual communication systems.