An investigation by Nikkei has revealed that generative AI has begun creating numerous anime images that closely imitate existing copyrighted characters, causing significant concern within Japan's anime industry. The inquiry discovered over 90,000 AI-produced images featuring characters from 13 internationally recognized anime series, with about 2,500 of these images showing strong resemblance to the original works.
Expansion of AI-Generated Anime Content
Advancements in AI have made considerable strides, capable of producing text, images, video, and audio by learning from extensive datasets. AI has been employed to generate and disseminate countless anime images on platforms such as civitai, pixai.art, and seaart.ai. By entering specific prompts, users can create and share images, but this practice risks infringing on copyright if the AI-generated images are too similar to existing works.
The investigation pointed out that the names of anime characters appeared in nearly 90% of the prompts used to create these images, suggesting intentional replication. For example, more than 1,200 images mimicking Pikachu from the Pokémon series were found, with some depicting Pikachu in controversial ways, such as holding weaponry or having altered physical features.
I was curious about this investigation. As a critic of the way tech companies train their AI models, I was eager to see how easy it is to mimic a popular character. Using Bing Image Creator and the simple prompt “Pikachu in a field,” I was given the image you see at the top of this report. While the biggest Pokémon fan may see the differences, to my non-fan eyes the AI image of Pikachu is a close resemblance to the real anime.
Legal and Industry Implications
Japan's legal stance on copyright infringement is determined by assessing “similarity” and “reliance,” where similarity examines the replication of unique expressions from the original, and reliance considers if the existing work has been referenced. Although no legal cases in Japan have yet tackled AI-generated images and copyright infringement directly, the swift production of these images poses a substantial issue.
In a related context, China's Guangzhou Internet Court ruled in February that an AI-generated image resembling Ultraman infringed copyright, exposing the susceptibility of Japanese cultural content. The unauthorized use of copyrighted material for training AI models is a growing issue, according to the “Anime Industry Report 2023” by The Association of Japanese Animations, which estimates the global anime market at about 3 trillion yen ($19.30 billion).
International Regulatory Approaches
Around the globe, regulatory actions are being introduced to combat the challenges presented by generative AI. In May, the European Union introduced the first comprehensive AI regulations, mandating transparency from AI providers. In the United States, the “fair use” doctrine permits the use of copyrighted works under specific conditions, with legality determined on an individual basis. Meanwhile, Japan's Agency for Cultural Affairs received over 20,000 public submissions expressing worries about unauthorized AI training.
