CD Projekt Red has reportedly used generative AI to modify Geralt’s Polish voice in The Witcher 3: Songs of the Past, matching new recordings to the actor’s performance before his stroke. I see a pattern of bringing the same actors back to their characters decades later.

It was a comment on the Witcher article that brought up the other side of this: we have to accept the passage of time. Voice actors get older. Their voices change, and so do their performances. A remake or remaster may need new recordings, just as a character’s design may need reconsidering. Keeping everything as it was is an awkward ambition for something also expected to feel new.

I do not mean to downplay the importance of bringing an actor back because their portrayal has become iconic. Having them recreate a famous scene is part of the appeal. But there is a difference between hearing that actor return and using AI alongside their performance to maintain the voice we remember. The latter can turn a celebrated portrayal into a requirement that neither the character nor the actor is allowed to depart from.

Stellaris comes to mind as a more clever use of AI text-to-speech. Its developers used generated voices for two artificial characters, Cetana and the Cyberpunk advisor, with the actors behind the voice models paid for each generated line, including future additions. The developer’s explanation was practical: the advisor could keep up with new mechanics instead of falling back to the default voice. A small addition to the script need not require another recording session, and the actor still gets paid for the use of their voice.

CDPR’s explanation draws a distinction between maintaining continuity within one game and preserving a voice for future installments. Its producer describes the voice matching as specific to the old and new recordings appearing together, suggesting an unadulterated performance may return in the future. The reported consent matters, and processing a new performance is different from generating one without the actor. Even so, permission to do it does not settle what we want from it. Bringing an actor back and keeping their performance fixed to the version we remember remain two different ambitions.

There is a limited time in which any actor can give us a particular performance. I do not think that needs to become a problem for AI to solve. A character can change, and another performance can mean something different. I want AI to bring an added benefit to the work. Keeping a franchise’s familiar voice unchanged should not become the condition for calling its portrayal authentic.