Meta has developed an open-source AI model called MusicGen that can generate music from text prompts and melody.
2
Prompts and melody
MusicGen is a transformer language model that can create 12 seconds of audio. Users also have the option to use reference audio to create music.
3
Training
MusicGen was trained on 20,000 hours of licensed music, including 10,000 tracks and 390,000 instrument-only tracks from Shutterstock and Pond5.
4
Availability
MusicGen's demo is available through Hugging Face's API. The code for MusicGen is also downloadable from GitHub but requires a GPU with 16GB of memory.
5
Better than rivals
Meta aims to address the lack of quality in AI music generators with MusicGen. The AI model seems to have an advantage over rivals like Google's MusicLM.