Quality not withstanding you’ve got to pay for access to the model or electricity to run your own local model, pay people to run the lines into the model and stitch them back into the game and pay people who speak the language to proof read the outputs to ensure it’s not giving you gibberish.
And if you’ve got voice lines now that’s a whole other can of worms of paying for TTS ai models, paying for audio mixing specialists, inserting the lines into the game, paying to once again have a speaker of the language QA test the output.
The electric costs aren’t nearly as high as people think. For huge datacenters, yes, but that’s because they’re processing requests for hundreds or thousands of users simultaneously. For a studio using it to translate lines for a single game, they could easily get away with doing it locally, and effectively for free. You can train your own local model on a consumer-grade PC without any issue, and it’ll still run just as fast as the big server farm-powered models.
My roommate has been playing with a bunch of different local AI models on his own PC for a couple years now. There’s been no discernable change to our electric bill. His PC draws more power playing an anime waifu gacha game than it does training/generating AI.
That still has associated costs my guy.
Quality not withstanding you’ve got to pay for access to the model or electricity to run your own local model, pay people to run the lines into the model and stitch them back into the game and pay people who speak the language to proof read the outputs to ensure it’s not giving you gibberish.
And if you’ve got voice lines now that’s a whole other can of worms of paying for TTS ai models, paying for audio mixing specialists, inserting the lines into the game, paying to once again have a speaker of the language QA test the output.
The electric costs aren’t nearly as high as people think. For huge datacenters, yes, but that’s because they’re processing requests for hundreds or thousands of users simultaneously. For a studio using it to translate lines for a single game, they could easily get away with doing it locally, and effectively for free. You can train your own local model on a consumer-grade PC without any issue, and it’ll still run just as fast as the big server farm-powered models.
My roommate has been playing with a bunch of different local AI models on his own PC for a couple years now. There’s been no discernable change to our electric bill. His PC draws more power playing an anime waifu gacha game than it does training/generating AI.