There is definitely that. I suspect the "overhead" that users pay for is less the processing horsepower and just straight up rental of the server time. Audialab claims - and I have no way of verifying this, obviously - that they're continuously refining and training their datasets so that would also be a reason there. If user usage is refining the models that would make sense. Of course, Microtonic's Patternarium functions in a similar way, with upvotes and downvotes influencing sound generation, and they don't charge more for it, so...take it all with a grain of salt, I guess.
Another simple reality is that their supporting server architecture might be highly dependent on Nvidia cards, this is a common problem right now for a lot of people to run the models locally. Sure the model itself would be fine on an iPhone, but you end up having to rewrite too much code and oh yeah, you are changing code constantly at the beginning.
Im eager for a developer to make a sound generator based on the point you made earlier of listening to samples you import and make variations upon that.
Heeeeeey well check out my latest video - a developer just did! It’s still in alpha but the technology exists and I think has some interesting ramifications.
That could be a lot of fun. Audialab does run its own discord and might have some resources for that already. Tensorpunk doesn't, to my knowledge, so there may definitely be a use for something like that.
Well, technically it's not AI, but ML - It's slightly more than " just" an algorithm, but slightly less than "classical artificial intelligence." But ML tends to get lumped in under AI studies, since it's in the same ballpark. Roger Penrose would never have complained about it, sure, but it's more adaptive based on input and dataset than a standard algorithm. More interestingly, you could use the same algortihm and get multiple different outputs depending on the training data, which is where the power of this lies - keep refining (or expanding) the data and you can get more interesting and complete output. It was also my second least-favorite CS class for my college degree, LOL. And that stuff has changed a LOT since I took that class [mumble] years ago.