Shanghai Court Orders AI Voice App to Pay HoYoverse ¥750,000
The decision treats a game publisher’s localized character assets as protectable interests when an AI service turns recognizable voices into a paid product. It leaves the separate fight over training data unanswered.
Listen to this story
The audio brief
Story brief
3 key pointsThe decision gives HoYoverse a commercial-law route to challenge AI voice clones tied to its game, even though the underlying performances came from voice actors. The app’s outputs were found nearly identical to original performances and were sold alongside Genshin Impact imagery and animations. The operator must pay CNY 750,000 and stop distributing the assets. The ruling is narrower than a training-data precedent:...
- 01
The app sold paid voice packs and character images designed to evoke 63 Genshin Impact characters.
- 02
The operator admitted using about one minute of official Genshin audio, plus other samples, without authorization for training.
- 03
The court recognized HoYoverse’s direct interest in protecting localized character assets, rejecting the defendant’s voice-actor-only rights argument.
A Shanghai court has ordered the operator of an AI voice-changing app to stop its service and pay HoYoverse CNY 750,000 after finding unfair competition in its cloning and sale of voices resembling 63 Genshin Impact characters.
The Shanghai Pudong New Area People’s Court ruled against an unnamed software company whose app offered paid voice resources and character images designed to make users sound like particular in-game figures. The yuan award has been reported at roughly $105,000 to $112,000, reflecting different currency conversions in coverage.
A paid product built on recognizable performances
The court’s finding was not limited to a generic voice effect. It found that the defendant used generative AI to replicate and sell the voices of 63 characters. The operator also used Genshin Impact images and story animations to market and sell the voice packs, conduct the court said showed an intent to capitalize on the game’s popularity.
According to reporting on the case, the operator admitted it had used about one minute of official Genshin Impact audio, along with other voice samples, without authorization to train its model. Forensic voiceprint analysis found the resulting outputs were nearly identical to the original performances.
The publisher interest prevailed, but the boundary is narrow
That contrast is central to the ruling. Rather than treating the dispute solely as one between a voice actor and an imitator, the court recognized the publisher’s interest in the package of localized character assets that makes a game character identifiable to players. The decision therefore gives HoYoverse a route to challenge commercial imitation even though the app’s outputs were based on vocal performances.
The case’s procedural path also matters. The defendant was ordered on June 30 to stop distributing the assets and pay damages. It briefly appealed, then later withdrew that appeal, leaving the order in place.
A ruling on commercial outputs, not a training-data answer
The reported significance of the case is specific: it has been described as Shanghai’s first unfair-competition case concerning AI recreation of voiceover work produced by another company. But it should not be read as a general answer to the legality of AI training on copyrighted works.
The ruling concerns the unauthorized reproduction and sale of character-like voices and related game materials through a paid service. Whether copyrighted material may lawfully be used as training data is a separate legal question that this decision does not resolve. For game publishers, voice performers, and AI toolmakers, that distinction draws a practical line: this case addresses what the service sold and how it was marketed, not every step in a model’s data pipeline.
Sources
- gamedeveloper.comMiHoYo wins $112,000 in Genshin Impact AI voice cloning lawsuit
- enduins.comHoYoverse Wins $105,000 in Genshin Impact AI Voice Cloning Lawsuit
Loading discussion...
Reader comments
Newest comments first. Replies stay oldest first.