MiHoYo has scored a legal victory in China after a Shanghai court ordered an unnamed generative AI developer to to pay ¥750,000 (about $112,000) after it found the company inappropriately reproduced and sold the voices of Genshin Impact.
This news landed last week at Chinese publication The Paper (via Automaton Media). According to The Paper (reviewed by machine translation), this is Shanghai's "first case" concerning "unfair competition" of using AI to recreate voiceover work produced by other companies. MiHoYo accused the defendant of not only using generative AI to replicate the voices of 63 Genshin Impact characters, but also used "images and story animations" from the game to advertise and sell its illicit voice packs to customers.
The court ruled that this "demonstrated an intent to capitalize on the game's popularity," as described by The Paper. The defendant was ordered on June 30 to cease distributing the assets and pay damages to MiHoYo. The unnamed company briefly appealed the ruling, but withdrew its appeal sometime in the last few months.
This ruling offers some guidance for how courts in China will assess the use of generative AI to replicate copyrighted material like the visual and audio likenesses of video game characters—relevant information for any developers looking to do business in the growing region.
It also speaks to a lesser-discussed risk of the spread of generative AI technology that exits in countries across the globe: the ability for third-party companies to create and sell tools that let users recreate copywritten material without permission or compensation. In 2023, voice actors Victoria Atkin and Tim Friedlander told Game Developer that they've seen their work recreated by players using third-party AI voice tools, who then create mods that insert them into other games. Some of those AI toolmakers, like ElevenLabs, also advertise their services to game developers.
It's worth noting that this latest ruling from China concerns the use of generative AI to replicate and sell copywritten material, and doesn't cover the still-fermenting legal questions around sourcing copywritten content as training data for AI tools.