MusicGen: A state-of-the-art controllable text-to-music model (by Meta Research)

• Generate conditional and unconditional music
• Generate outpaintings from a prompt and a NendoTrack
• Use a NendoTrack as melody conditioning to generate music
• Use custom finetuned musicgen models

For a list of community finetunes check out community_finetunes.json. If you want to contribute one to the list, open a PR or contact us via Twitter or Discord!

Since we depend on audiocraft, please make sure that you fulfill their requirements. You need Pytorch 2.0.0 or higher, which can be installed via

pip install "torch>=2.0"

Note: On Mac OSX, the instructions for installing pytorch differ. Please refer to the pytorch installation instructions. For all other problems please refer to the audiocraft repository.

To run the plugin you also need to install audiocraft by Meta AI Research, run:

pip install git+https://github.com/facebookresearch/audiocraft

1. Install Nendo
2. pip install nendo-plugin-musicgen

Take a look at a basic usage example below. For more detailed information, please refer to the documentation.

For more advanced examples, check out the examples folder. or try it in colab:

from nendo import Nendo, NendoConfig

nd = Nendo(config=NendoConfig(plugins=["nendo_plugin_musicgen"]))

# load track
track = nd.library.add_track(file_path='/path/to/track.mp3')

# run musicgen with custom model
generated_collection = nd.plugins.musicgen(
    track=track,
    n_samples=5,
    prompt="janelle monae, rnb, funky, fast, futuristic",
    bpm=116,
    key="C",
    scale="Major",
    duration=30,
    conditioning_length=10
)
generated_collection[0].play()

Visit our docs to learn all about how to contribute to Nendo: Contributing

Nendo: MIT License

AudioCraft: MIT License

Pretrained models: The weights are released under the CC-BY-NC 4.0 license