<p>This article addresses uncertainties over authorship in the age of generative AI by developing the theoretical underpinnings of a systematic approach to attributing authorship in visual art practices involving generative AI, which build on the workings of multiple agents and technologies. By using an analytic philosophical methodology to analyze the practices and key concepts under discussion it is clarified what is meant by authorship in these practices and what kind of works are at stake. As this analysis finds, authorship becomes complicated in these practices not just because of the non-deterministic workings of generative AI systems, but because the works under discussion can be viewed on the one hand as the outputs of generative AI systems and on the other as works of AI art, including kinds referred to here as “synthetic images”. Importantly, authorship in the latter case requires intentionality and autonomy from agents to both initiate the creative process and ratify, or evaluate, the work as theirs, something that only humans can do at this point. Nevertheless, the dependence of synthetic images on AI-generated images means that in attributing authorship in the practices under discussion, we ought to account for the dual identity of these items and, accordingly, the different forms of authorship and contributions attached to them. This dual approach, it is demonstrated, provides the tools to assess when the potentially many and varied figures involved in these practices should count as authors, or other kinds of contributors, and facilitates more rewarding appreciation of the artworks.</p>