The latest OpenAI’s version of ChatGPT, an AI-enabled chatbot allows users to generate images of themselves, their pets, films or memes in the distinct “Ghibli” style of Studio Ghibli, the acclaimed Japanese animation studio. Notably, not all users are on board with the “Ghiblification” of images by OpenAI’s new tool. This post looks at the legal issues surrounding this new ChatGPT feature.
Blurring the line between homage and outright infringement
Legally speaking, artistic styles per se may not be protected under copyright law. However, specific characters or scenes like Totoro or a frame from Princess Mononoke are legally protected.
However, when ChatGPT allows a user to generate an image in an ‘Ghibli-style,’ OpenAI may likely be considered as trading off the goodwill of Ghibli’s trademarks (i.e. using Ghibli’s identifiable style) which may lead to a likelihood of confusion among consumers that such a function is indeed endorsed or licensed by Studio Ghibli.
Therefore, if ChatGPT produces a “Ghibli”-fied image that may be too close to the originals, such image may be subject to a potential lawsuit.
Walking on the thin line of copyright law
To date, there isn’t any legal precedent to determine if OpenAI contravenes the copyright law.
AI companies like OpenAI for training data for its models have used the “fair use doctrine” under copyright law as a legal protection. The “fair use doctrine” is a legal grey zone. The doctrine allows limited access to copyrighted materials without prior permission (e.g. quotations, research, teaching, news reporting, and other non-infringing uses).
However, OpenAI is already facing several lawsuits over its approach to scrapping the internet for training data for its models, including plenty of copyrighted material. In 2023, The New York Times, a news company sued OpenAI that the AI company infringed its copyright by using its material without permission to train its AI models.
To summarise, it may be likely that OpenAI’s model was trained on Studio Ghibli’s work. That in turn may raise the question if OpenAI has obtained a license or permission to do such training or not? Therefore, if we may look at the output of generative AI and see identical elements or substantially similar elements in that output, and such type of use was happening without consent and compensation, it may be problematic.
Time to rewrite the rulebook?
Copyright laws, developed centuries for a pre-AI world will need an overhaul. Instead of relying on the courts to interpret the existing copyright law, lawmakers worldwide must act quickly to discuss with a view to modernise copyright law for the age of AI. Should AI creation be copyrighted? What counts as fair use when machines exploit such original work?
A potential compromised scenario may include where artists agree to a system where the AI developers will compensate them as the copyright holders, and credit them where their content is used (in producing AI output).
AI developers should be transparent on data sources such as disclosing how they source their data (e.g. public domain, licensed content, or under fair use). Or they may likely face further legal challenges from copyright owners.
Final thoughts
When Studio Ghibli Co-Founder Hayao Miyazaki saw an AI demo in 2016, he was ‘utterly disgusted’ by it, as captured in documentary footage. He had warned in the past that AI could become ‘an insult to life itself’ if it loses sight of the human soul.
Can we prove him wrong? “Ghiblification” may be just the beginning, as we foresee future clashes between copyright law and AI. The challenge lies in whether we can strike a balance between rewarding innovation and upholding artistic integrity. Lawmakers need to act swiftly to provide regulatory clarity on the fair use of AI for the benefit of all.
This article was first published on e27.