A Study on the Copyrightability of AI-Generated ‘Text-to-Image’: A Classification-Based Analysis

Abstract

The advent of text-to-image generative AI models such as DALL-E 2, Stable Diffusion, and Dreamfusion has brought the convergence of artificial intelligence and art to contemporary aesthetic discourse. At the same time, these developments present unprecedented challenges to traditional copyright law. The highly automated nature of AI-generated content often renders conventional criteria for defining "works" inadequate. Additionally, this issue is influenced by many factors, including cultural contexts, social structures, economic policies, and legal systems, contributing to its complexity and underscoring the necessity for a global consensus. This paper builds on the legal framework governing the copyright protection of photographic works to propose differentiated legal protection strategies for distinct categories of text-to-image generative content: Original Works, Derivative Collaborative Works, and Automated Outputs. Specifically, it advocates for a classification-based approach to protection, guided by the originality, degree of creative effort, and aesthetic value inherent in the generated content. Furthermore, the paper recommends introducing aesthetic evaluation criteria and establishing formal appraisal mechanisms to refine standards for copyright eligibility. This framework aims to balance the dual imperatives of robust copyright protection and the broader public interest by addressing the distinctive characteristics of generative AI creations. Ultimately, it provides new theoretical insights and practical guidance for the legal regulation of generative AI content, offering a pathway for harmonizing innovation with legal and societal norms.