Extension of the AI Pastiche Dataset and Evaluation of Generative Models Progress on the Problem of Imitating Painting Styles

This work presents an extension of the AI-Pastiche dataset aimed at evaluating the current capabilities of generative artificial intelligence models in replicating historical artistic styles in the production of images of paintings from textual prompts. The extension follows the same data collection procedures and evaluation methodology adopted in the original AI-Pastiche study, ensuring full compatibility and comparability with the previously collected data. Four of the models included in the extension are more recent versions of models already examined in the original work, enabling a direct comparison between successive generations of systems belonging to the same model families. The comparative analysis highlights a general improvement in the ability of modern models to generate visually credible images that adhere more closely to the requested prompts. In particular, the results indicate progress in the representation of artistic styles, with greater fidelity to the specified pictorial techniques and a reduced tendency toward hyperrealistic or historically inconsistent outputs. Improvements were also observed in subject representation and in the reduction of visual artifacts. Despite these advances, the results confirm that generating historically and artistically credible paintings remains a challenging task. Complex scenes and contexts that require a deeper understanding of stylistic conventions and historical coherence still expose limitations in current generative models. Overall, the extended AI-Pastiche dataset provides an updated resource for analyzing the evolution of generative image models in the artistic domain. By enabling systematic comparisons across model generations, it supports the study of progress and persistent limitations in artificial artistic image generation, while also contributing to the broader discussion on the relationship between artificial generation and human creativity.