Who owns AI art? AI-generated Ghibli selfies spark legal debate

What began as a nostalgic craze — turning selfies into illustrations styled after Japan's much beloved Studio Ghibli — quickly went viral, drawing millions of users to ChatGPT’s new image generation feature.

With a simple photo upload and a text prompt like “turn this selfie into a Ghibli-style image,” OpenAI’s GPT-4o model can transform the image into one that mimics the studio’s distinctive works within seconds, based on the vast variety of image styles it has been trained on.

But behind the charm of these cutesy animated selfies lies a serious legal debate: Who owns the copyright to AI-generated images?

OpenAI has clearly benefited from the viral popularity of its image generation feature. Since the launch of GPT-4o on March 25, ChatGPT’s user base has soared past 500 million, with more than 700 million Ghibli-style images generated through the app to date.

In South Korea, ChatGPT topped the weekly mobile app download charts, with 2.72 million new installations recorded between March 31 and April 6, according to IGAWorks’ Mobile Index. Over the past year, paid subscribers in Korea have increased by roughly 30 percent, with 14 percent being new users with no prior payment history, according to a report by NongHyup Bank.

As AI-generated content becomes more visible, other platforms are also gaining traction, such as Midjourney, which generates illustrations based on image descriptions, and Suno AI, which can compose music using just lyrics and genre inputs.

Studio Ghibli has remained silent on the trend. However, the debate has resurfaced past comments by the studio’s co-founder and legendary animator Hayao Miyazaki, who once referred to such images as “an insult to life itself.”

“I am utterly disgusted,” he said in a 2016 video, reacting to a computer-generated animation of a grotesque creature.

At the center of the legal debate surrounding generative AI are questions about both the source of training data and the ownership of AI-assisted works.

Most AI models are trained on vast datasets scraped from the internet — often including copyrighted material used without permission. This has sparked concern over whether content generated from such training could be considered a byproduct of infringement.

At the same time, the growing use of AI in creative work is raising new questions about whether content produced with the help of AI tools can be protected under copyright law, and how much human input is required to meet the legal standard for originality.

In the case of GPT-4o’s Ghibli-style illustrations, questions have emerged over whether the AI company has infringed on the studio’s copyright — and whether users can claim ownership of the images they generate using the model.

Last week, Japan’s Ministry of Education, Culture, Sports, Science and Technology addressed the issue, stating that AI-generated images that merely resemble Ghibli’s artistic style are unlikely to violate copyright law.

“If the similarity lies only in style or concept, it does not amount to copyright infringement. Ultimately, however, it is a matter for the courts to decide,” said Hirohiko Nakahara, a strategic officer at the ministry.

Similar uncertainty extends to AI-generated music, writing and other creative content.

The global music industry is likewise grappling with unauthorized use of original songs for AI training, as well as the rerelease of altered tracks using those materials.

According to the Korea Copyright Protection Agency, between March 28 and April 11 last year, 1,647 AI-generated cover songs were uploaded to YouTube using the voices of 254 singers without permission. Of those, 92.9 percent involved songs originally released in Korea.

In Korea, a team of researchers has raised a critical question: Can AI-generated works still be regarded as “purely mechanical outputs” when they involve a meaningful degree of human creativity?

In a recent study published in Telematics and Informatics, titled “Who owns AI-generated artwork? Revisiting the work of generative AI based on human-AI co-creation,” researchers argue that the legal definition of human-AI co-created works will be essential in determining how copyright is applied in the era of generative AI.

“Even when an image is generated by AI, it cannot be easily dismissed as a purely mechanical output if there is meaningful human creative input involved,” said Hwang Yo-han, the professor at Jeonbuk National University who led the research team.

“A major challenge going forward will be establishing how the concept of human-AI co-creation should be defined and protected under the law.”

The industry has yet to reach a clear consensus on authorship or ownership, and different countries are taking different approaches to regulation.

In 2023, the US Copyright Office revoked the copyright registration for AI-generated images in the graphic novel "Zarya of the Dawn." Despite the applicant having written more than 1,500 prompts, the office ruled the applicant could not be recognized as the author of the resulting images, stating that the final outputs could not be sufficiently controlled to qualify for copyright protection.

In contrast, a 2023 ruling by the Beijing Internet Court in the Li v. Liu case found that AI-generated images could qualify for copyright protection. The court concluded that the user’s act of selecting parameters and entering prompts involved a sufficient level of creativity, viewing the AI software as a tool, akin to tools that draw lines and apply color at the user’s direction.

In Korea, efforts are underway to establish clear guidelines for AI content generation. Led by the Ministry of Culture, Sports and Tourism, the government is planning to revise the Copyright Act to require disclosure of datasets used in AI training and ban unauthorized removal of labels identifying content as AI-generated

Meanwhile, Korea’s Unfair Competition Prevention Act already prohibits the unauthorized use of content created through another party’s investment and effort, particularly when such use infringes on the original creator’s economic interests.