Deepfakes in Visual Art: Differentiating AI-Generated Art From Human Art Using Convolutional Neural Networks (CNN)

Abstract

As AI technology evolves, seeing is not believing. The boundary between human and machine creativity is increasingly blurred, presenting challenges for the art industry. This is more pronounced nowadays as advancements in AI technology make it increasingly easy to create highly realistic synthetic art. This study explores the use of Convolutional Neural Networks (CNNs) to differentiate AI-generated art from human-created art. By employing Error Level Analysis (ELA), an image forensic technique for detecting fake and real images, this study develops a robust CNN classifier. Using the AI-ArtBench dataset, the optimal model achieves a 99% classification accuracy, even when tested on art from a different generative model. While AI-image detection remains a “cat and mouse” pursuit due to advancements in generative AI, the findings of this study highlight that there are clear, discriminable differences between AI-generated and human-created art. The implications of this research extend beyond academic inquiry. They offer support for artists, collectors, curators, and policymakers as they navigate the complexities of AI’s expanding availability and address the evolving role of AI in the art world. It sets the groundwork for application within fields faced with the same or similar challenges.