MusicGen (AudioCraft) 🧠 Model Open source

What it is

MusicGen is Meta's text-to-music model, part of the AudioCraft research library. It generates instrumental music from a text description, optionally guided by a melody you hum or upload. AudioCraft also includes MusicGen Style, MAGNeT and JASCO, a model that takes chords, melodies and drums as conditions. MusicGen was first released in 2023 and is now older than the commercial leaders. We found no newer Meta music model in the repository.

Key features

• Text-to-music and melody-conditioned generation
• Several model sizes, from small to large, plus stereo variants
• Instrumental only, no vocals
• MusicGen Style and JASCO add style and chord conditioning
• Python library with training code for research

Where you can use it

Install AudioCraft with pip and run it locally, or try community demos on Hugging Face. We found no creator app in our directory that documents MusicGen as a built-in model.

Pricing and rights

The AudioCraft code is MIT-licensed, but the model weights are CC-BY-NC 4.0, which rules out commercial use. You can learn, experiment and prototype, but you cannot put MusicGen output into monetised videos or client work under that licence.

Who it is for

Researchers, students and hobbyists exploring music generation, and developers prototyping locally before choosing a commercial model.

Verdict

MusicGen was an influential open model and is still easy to run. The non-commercial weight licence, the lack of vocals and its age make it a poor fit for creators who publish. ACE-Step or Stable Audio's open weights are better open choices today.