Please listen to and rate this song: https://soundcloud.com/steve-sokolowski-2/12-rythmos-bay
This song was 100% predicted by AI models. However, as different models were used and the same stems were used to predict multiple sequences, their predictions were stitched together in an external editor. No non-AI sounds were added to the mix.
This song uses a new method I discovered of chaining AI models together to generate very complex prompts and lyrics, which I'll reveal after the poll closes so as not to bias the results.
Thanks for voting! There is one final song, which I think is even better, at /SteveSokolowski/how-good-is-this-new-song . The help here has assisted greatly in refining my prompts, and I'll be posting a 8-page document detailing how I did it after I get feedback on the last song.
1,000
1.00If @TheAllMemeingEye is willing to weigh in, as he rated this song excellent (anyone else can too), I'd be curious as to whether the following demo is better or worse than "Rythmos Bay:"
https://drive.google.com/file/d/1ZpjF1JAgy7PQuL6ZYqJ_cKQUDeJOEmu5/view?usp=drive_link
There are some errors and I don't like the ending yet, but in this song, I changed my "coding" in the models to improve the vocals so that I was able to get a full chior as backing.
I want to publish my method somewhere because I definitely think these models are much more capable than people know, but I've found that my own ear is the limiting factor - I didn't pay attention to vocals as much in prior songs, for example because I personally didn't care about them.
Wow, how is this vote so polarized?
The previous songs were voted around a 4 and the opinions were consistent. I wonder if there are people rating the song because they don't like the genre, rather than its technical aspects?
@paphnutius Because on the Internet, people are always willing to downvote stuff without listening to it or reading it, but nobody ever says anything positive about people unless it's actually positive.
I think I probably should make another poll about this song, which I might do. This itself is a useful poll - someone clearly thinks this is a pop song as good as anything else. But I am also interested in determining whether my prompting method has resolved the common issues that people associate with AI music, like simple instrumentation, a lack of stereo width, and "compression artifacts."
I essentially want not only to determine if it's a good song, but also whether even if this isn't people's cup of tea, this music production method can now be used interchangeably with traditional methods. I think I have a way to do this - creating a market for it.