I believe, at some point (early last year, if my brain isn’t doing a funny), the industry generally configured their models to use this glossy cartoon style, if the user doesn’t specify anything else, because it’s a style the models can do fairly well.

Their largest training set is humans, but mistakes on humans look horrid, due to Uncanny Valley, so the cartoon style avoids that. It also requires relatively few details that the model can get wrong or that the user would have to correct in their prompt (like perspective, light angle etc.).

Well, and this specific cartoon style allows them to still keep lots of detailing where mistakes don’t matter, like on the hair, so that it doesn’t look entirely simplistic…

I don’t know, if they get many listens, but 1½ weeks ago, this article made the rounds from Deezer, saying that they do get a ton of AI-generated uploads: https://newsroom-deezer.com/2026/07/ai-music-exceeds-50-percent-daily-uploads-deezer/
Granted, this is definitely also a marketing piece for Deezer to talk about how great their AI detection is, but I also don’t think they would claim there to be more AI-generated songs than there really are, as they don’t gain much from lying here.

At this point, other streaming services are also actively trying to curb back AI-generated songs, since they mostly just take up extra storage space. We’ve had such a massive abundance of songs before already, that you really just needed good curation, not more of them.

In that sense, I do assume that the AI-generated music is not seeing as many listens as the uploaders may have hoped for. And maybe that also means that they do not hurt artists quite as much in the end.
I imagine, these uploaders just hope for one song to do well, to not be detected as AI-generated etc., and for that song to pay for dozens more generated songs.

There’s folks that spam music streaming services with thousands of AI-generated songs in hopes of making a few bucks.

So, you don’t actually need it to be hugely popular for it to hurt artists…

we believe they want to know when they sound inauthentic.

I mean, that’s everyone on LinkedIn, but okay.