Artificial Intelligence (AI) has transformed the creative process in digital media art, yet existing generative models often struggle to balance realism, interactivity, and creative diversity. This study proposes a hybrid framework that integrates generative adversarial networks (GANs), reinforcement learning (RL), and particle swarm optimization (PSO) to generate dynamic and interactive artworks that evolve in real time based on audience engagement. The core innovation lies in combining GANs’ image realism with RL’s adaptive responsiveness, while PSO optimizes hyperparameters for faster convergence and higher generative quality. Style transfer is applied prior to principal component analysis (PCA) to enhance stylistic diversity before dimensionality reduction. The model training using the WikiArt dataset of over 80,000 paintings, the proposed system demonstrates superior performance compared to established benchmarks, including StyleGAN2, VQGAN + CLIP, and diffusion models. To evaluate perceptual quality and user engagement, a structured user study incorporating semi-structured interviews and a questionnaire-based survey was conducted with 200 participants, capturing quantitative and qualitative assessments of realism, creativity, interactivity, and overall user satisfaction toward the AI-generated artworks. Quantitative evaluation using Fréchet inception distance (FID), inception score (IS), and CLIP-score, along with user studies, shows significantly improved realism (M = 4.6, SD = 0.8), creativity (M = 4.5, SD = 0.7), and interactivity (M = 4.4, SD = 0.6) on a 5-point Likert scale. This work contributes a reproducible framework for computational creativity that bridges artificial intelligence and digital media art, advancing human–AI co-creation and interactive aesthetic design.
Ji et al. (Fri,) studied this question.