Meta takes on Google’s MusicLM AI with MusicGen, here’s how it works

The Audiocraft research team at Meta has recently launched MusicGen, an open-source deep-learning language model.

MusicGen can produce new music based on text prompts and can even be aligned with an existing song. The model is similar to Google’s MusicLM, trained on 20,000 hours of licensed music. It can also take melody as input and complete it with its musical creativity.

On Facebook’s Hugging Face AI site, there is a demo that allows you to describe your preferred music. You can select from a few examples such as “an 80s driving pop song with heavy drums and synth pads in the background.” Afterwards, you can “condition” your selection with a song up to 30 seconds long.