Using Image Similarity Metrics to Discriminate Between DALL-E Generated Art and Original Art

Abstract

The rise of image-generating artificial intelligence (AI) tools like OpenAI’s DALL-E has changed the way art is made. It brings up important questions about originality, authorship, and ethical implications. I explore the originality of AI generated art through a quantitative similarity analysis using Bhattacharyya distance and Euclidean distance to measure color and structural similarity between AI outputs and their reference artworks. I analyze three distinct prompt variations—one including the artist’s name, another with a detailed description but no artist reference, and a third requesting a reinterpretation rather than replication of an original painting, namely the Mona Lisa. I find that AI generated images exhibit varying degrees of similarity depending on prompt specificity. The metrics found that mentioning the artist’s name in the prompt resulted in more similar outputs than when asked for a direct reinterpretation. Similarity metrics indicate that AI generated outputs tend to resemble each other more closely than they do the original painting, implying that AI models operate based on learned visual patterns rather than direct replication. The study emphasizes how important it is to have explicit ethical guidelines and legal frameworks to be able to regulate AI’s influence in the artistic domain.