As digital marketing ecosystems expand, advertising creative production demands have risen significantly alongside requirements for visual coherence and style control. Traditional manual workflows struggle to meet large-scale customisation needs. The research addresses creative generation and style consistency challenges through coordinated modelling of semantic interpretation, creative expression, and stylistic regulation, balancing automation with visual identity stability in commercial image creation.
The study introduces generative frameworks leveraging Diffusion Models and Generative Adversarial Networks for automated advertising synthesis. Text-driven mechanisms map prompt semantics into visual representations through Transformer-based encoders, extracting features transformed into conditional inputs controlling image composition and layout. Multimodal modelling uses contrastive learning, projecting text and images into unified semantic spaces, while brand tonality modelling constructs quantifiable attribute representations embedded through style vectors and adaptive normalisation parameters.
Platform implementation incorporates a modular architecture trained on 18,000 advertising images from 42 brand projects spanning multiple industries. Template library construction processed 1,120 image groups with structured layout annotation. Performance validation demonstrated prompt parsing latency of 85.3ms, image generation duration of 683.2ms, and semantic alignment accuracy of 87.4%, confirming framework feasibility for automated content creation, maintaining compositional and stylistic coordination across brand campaigns.
Contributing to this research is Jing Zheng, holding a Master of Science in Mathematics from New York University’s Courant Institute and a Bachelor of Science in Mathematics from the University of Texas at Austin. Zheng’s background reflects quantitative academic training together with professional software engineering roles. Technical skills include Java, Python, and SQL, as well as frameworks such as Spring Boot, Django, and Vue.
Since August 2022, Jing Zheng has worked as a Software Engineer and has developed web and mobile applications that make it easier for advertisers to create effective ads and promote their products. She also led the development of machine learning models that optimized ad performance recommendations for small and medium advertisers, reducing zero-outcome ads. In collaboration with Generative AI teams, Zheng developed and launched the AI-generated media feature, which empowers SMBs to generate high-quality creatives through prompt-based image generation. She further engineered and shipped a strike-based enforcement policy and developed predictive machine learning models, which led to meaningful reductions in fraud, abuse, and low-quality advertisers within messaging ads.
The study demonstrates a prompt-driven and style-coordinated framework that supports automated advertising creation with stable performance across deployment scenarios. By integrating structured template referencing and brand-specific style embeddings, the system enables compositional and stylistic coordination in generated outputs. Empirical validation confirmed consistent response characteristics and high semantic alignment accuracy, indicating that the platform meets diverse production needs and offers a viable foundation for future extensions in intelligent advertising generation.