Jang Se-Min of the AI Times reports that the company Nara Knowledge Information received, in South Korea, a copyright registration for its AI-generated film titled “AI Suro’s Wife.” The registration was filed and completed on Dec. 29, 2023 for an “edited work” based on the human editing of the AI generated film and images.
Nara described its editing or involvement in the creation of the final film:
(1) ensuring consistency in the images from prompt engineering, so that they were not just random, which involved “the process of selecting text, pictures, voices, etc. at each stage and reaching the final result can itself be close to creation,”
(2) edits through “Photoshop, out-painting and in-painting were repeated dozens of times, even the parts that AI could not capture were enlarged and depicted,”
(3) editing the images in Midjourney and Stable Diffusion,
(4) incorporating images into the video, and
(5) relying on their own small language model for some elements, such as to help incorporate the Korean song “Surorbun” into the work.
According to the AI Times, under South Korea’s 2023 Copyright Registration Review Handbook, AI generated works can qualify for copyright based on a human’s selection and arrangement of elements in editing the AI generated works:
‘If elements of traditional copyright are implemented by artificial intelligence technology, copyright is not recognized, but although humans have modified the artificial intelligence product in selection, arrangement, etc., copyright is recognized to a limited extent.’
South Korea’s 2023 Copyright Registration Review Handbook
I haven’t been able to independently review South Korea’s AI Copyright Guidebook, which was announced on Dec. 27, 2023. Some reports described that copyright would be allowed only if a work “demonstrably convey human thoughts and emotions.”
If the AI Times’ article is accurate, the approach of South Korea to the copyrightability of AI generated works sounds similar to the approach in a recent decision by the Beijing Internet Court for an AI-generated image that was created and refined through repeated prompts of a human creator.
It’s also similar to the approach that the U.S. copyright law also recognizes, in my view, notwithstanding the contrary view taken by the U.S. Copyright Office in several rejections of registration to AI generated images. I believe the Office’s position is, with all respect, wrong. It has yet to be reviewed by the federal courts.
I contend that U.S. copyright law, properly understood under the Copyright Clause and Supreme Court precedents, recognizes authorship based on a minimally creative selection and arrangement of elements by a human creator, but any copyright for such a minimally creative selection and arrangement is thin, protecting against only identical copies. If this principle sounds familiar, it should: it comes straight out of the Supreme Court’s seminal decision about originality in Feist.
I argue that this fundamental principle of U.S. copyright law applies to AI generated works. To learn more about my analysis:
Edward Lee, Prompting Progress: Authorship in the Age of AI, Florida Law Review (forthcoming)
Edward Lee, Comment of Professor Edward Lee to Artificial Intelligence Study by The United States Copyright Office (Oct. 30, 2023)
Edward Lee, A terrible decision on AI-mage images hurts creators, Washington Post (April 27, 2023)