Not to be outdone by Google, Meta released its own AI-powered music generator – and, unlike Google, made it open source.
It's called MusicGen, and it's a music creation tool by Meta. A demonstration of the project you can find here. The company's AI tool can turn a text description (eg, “An '80s pop song with lots of drums and synthesizers in the background”) into 12-second files. MusicGen can optionally be “directed” with some reference sound, such as an existing song, so it will try to follow both the description and the melody.
Meta reports that MusicGen was trained with 20.000 hours of music, 10.000 “high quality” licensed music tracks, 390.000 instrumental tracks from ShutterStock and Pond5, a largecase multimedia.
The company has not provided the algorithm it used to train the model, but it has made available pre-trained models that can be run on any device with the appropriate hardware (GPU with around 16GB of memory).
Does MusicGen work? Its songs are melodic, at least for basic commands like “ambient chiptunes music,” and it seems to be on par (if not slightly better) than Google's own music maker, the musiclm. Of course, neither of them will win any music awards.