AECO-SC StyleGAN: A Cross-Platform GAN Framework for Dynamic Advertising Creative Generation

Abstract

Dynamic advertising (ad) requires personalized, engaging content across multiple platforms. Traditional approaches struggle with scalability and cross-platform adaptation. Leveraging deep learning (DL), particularly Generative Adversarial Networks (GANs), offers the potential to automate and optimize ad creative generation with higher precision and contextual adaptability. This research aims to develop a DL framework that dynamically generates and optimizes advertising creatives—leveraging Adaptive Elephant Clan Optimizer with a Spatially Conditioned StyleGAN (AECO-SC StyleGAN) for dynamic cross-platform advertisement creative generation. Adaptive Elephant Clan Optimizer (AECO) dynamically adjusts training hyperparameters to improve model convergence, while Spatially Conditioned StyleGAN (SC-StyleGAN) generates platform-specific ad creatives by incorporating spatial constraints for contextual alignment. Our system is trained on the Ad ImageNet dataset, which includes 9,003 ad samples with paired images and promotional text from platforms like Facebook and Instagram. All data were resized to 256×256, normalized, and tokenized for training. Using Python, the model demonstrates superior performance in creative generation and engagement prediction. The proposed AECO-SC StyleGAN model achieved an NDCG of 0.61, an accuracy of 98.48%, and a weighted F1-score of 98.5%, outperforming prior approaches such as VGG + Layout + NIMA (NDCG 0.22) and XCEPTION (accuracy 98.27%, F1-score 98.2%). These results highlight the effectiveness of integrating adaptive optimization and spatial conditioning in generating high-quality, context-aware advertising creatives, offer a scalable and automated solution for cross-platform digital marketing.