Reconstructing Multi-Style Features in Brand Posters via StyleGAN and Generating a Unified Template Structure

To address multi-style feature coupling, inconsistent template structures, and low design efficiency in traditional brand poster design, this paper proposes a StyleGAN-based method for reconstructing multi-style features and generating unified template structures. A multidimensional feature dataset containing more than 12,000 brand posters is first constructed, covering color systems, layout structures, graphic elements, and font styles. The StyleGAN2-ADA model is then customized by optimizing the mapping network and style injection module. A hierarchical control strategy for style feature vectors maps core features, including color, layout, graphics, and typography, to different StyleGAN layers, improving single-style feature control accuracy by 42% and enabling precise decoupled style reconstruction. A unified three-tier template architecture of “feature reconstruction-template generation-dynamic adaptation” is further designed, using StyleGAN to generate basic template frameworks. Brand VI guidelines are integrated to constrain style feature combinations, ensuring layout consistency and stable core-element positioning across posters with different visual styles. The method provides an efficient generative-design approach for structured brand visual communication.