“Webtoons” and Generative A.I. in South Korea – Is Artistic Style Copyrightable?

I came across an interesting situation unfolding in South Korea regarding the development of image generating A.I. for web comics (commonly known as “webtoons” in Korea). As a preface, I would like to highlight that I do not know much about webtoons nor about industry all that well. I tried my best translating Korean to English, but I must admit that some sentences doesn’t flow very well, especially with all the jargons. If anyone is more versed in the area, I would be more than happy to try to clarify the matter.

The situation pertains to a relatively new start-up A.I. company called Onoma AI that develops A.I. tool which provides services that creates scenes for webtoons. Onoma AI was selected as part of Deeptech Incubator Project for Startup (DIPS* 1000), a start-up venture funding program by the Korean Institute of Startup and Entrepreneurship Development and Ministry of SMEs and Startups.¹ In an article, CEO of Onoma AI, Song Min, noted that creative A.I. is not the enemy of webtoon and that change of perception towards generative A.I. is necessary. Song Min further noted that generative AI should not be feared as replacement of human [artists] but many use them as a tool that supports creative work.²

Allegations came from an artist name JaGae, who is publishing a webtoon series on one of the top search engines in South Korea, Naver. From what I hear, Naver is one of the largest, and one of the best platforms to publish webtoons and is regarded very highly in the industry.

JaGae, on X, noted that they received an email notifying them of an A.I. website which may have used their drawings. They clarify that permission to use their work for A.I. learning never occurred,³ then went on to compare use of their work for A.I. learning to theft.⁴ Following this X post, another Webtoon artist, Jeokbal, openly prohibited use of their literary, and artistic work for A.I. learning, noting that this directly relates to livelihood of artists.⁵

It seems that allegations revolving around the use of copyrighted materials to train Onoma AI’s program turned out to be true, and the company issued a statement of apology the following day:

“Recently, we received several concerns regarding learning of certain artist’s style (emphasis added) without their permission… …In the beginning stage of the development, Onoma AI used external models for internal testing purpose, and exchanged feedback that certain artists style can be seen in the external models.

Not infringing artists’ copyright is an important vision of our company, and therefore we decided not to use external models. However, while beta versions circulated, models with external inputs were released… … any models that had external inputs has been taken down. Currently our service only produces images that use self-image-model.

The purpose of our program is to reduce repetitive tasks for artists and enable them to focus on creative endeavors… …Currently, all data required for image-learning are obtained using copyright free illustrations, tuned through internal retoucher which adds webtoon style.”⁶

I omitted some texts that I deemed unnecessary for understanding the context, but if you are interested, feel free to ask me to clarify or add details.

In an article published in June of 2024, it is purported that Onoma AI used over 10 million illustration images and used webtoon data from partners.⁷ However, it seems that, after Onoma AI’s own admission, copyrighted materials were indeed used to train the A.I. without permission.

Now, this case interested me given how the situation played out. First of all, I am not quite sure how the artist, JaGae, found out that their image was used for A.I. learning. In the Onoma AI’s statement, it discusses “certain artist’s style” which further makes me question whether or not a style of art can be copyrighted in some way (hence the title). I am not sure whether or not the artist found examples of their work on Onoma AI’s website somehow, or the final result blatantly resembled their work. However, when reading the statement, I thought of certain drawing styles such as Disney animations or Marvel comics. As someone who rarely reads comic books, I can probably tell the artistic style between Disney and Marvel apart, but I don’t think I could tell Marvel and DC apart. It is my preliminary thought that such artistic style cannot qualify for copyright, which made the statement from Onoma AI to be quite peculiar.

Secondly, it seems that Onoma AI only used copyrighted materials without permission for beta versions for internal testing purposes. That said JaGae further questions how Onoma AI can be certain that other copyrighted materials are not “mixed in” with the model even at the current stage.⁸ Indeed, it makes me wonder if the remnants of the materials used at the early developmental stage are still used to produce final work, particularly when the program reached commercial level. Interestingly, Onoma AI released an open-source version of one of their A.I. models,⁹ and while it is not completely clear if the open-source model is the same one being commercially serviced, it makes me wonder if Onoma AI can argue public interest based on this.

At the end of the day, I found it quite surprising that Onoma AI admitted to using copyrighted material without permission and issued an apology and would love to hear other’s opinions on this matter.