Google has developed a new AI that can create minutes-long songs that can be developed by inputting text. It's called MusicLM, and while you can't use it at the time of writing, you can hear some samples here.
Some AI generations are not the highest quality, but after listening to these, I could see these being composed by a human. There are different modes, them being audio generation from captions, long generation, and story mode, which is a given script.mainly it mostly uses instruments, the AI can also (sort of) replicate human vocals. The vocals sound grainy and staticky, but you can still tell that they are meant to be human.
If you are wondering how Google manages to do this, they have released research papers explaining how they get these results.
The last part is, in my opinion, the coolest part of this demo. They had people hum, whistle, or play an instrument to have the AI swap the instrument. For example, they had someone hum "Turkey In the Straw" and it changed it for a Jazz with Saxophone piece.
While we could get the tech for making this piece, so far the only people who can use this are the people developing the AI.
