MusicGen
Generate instrumental sketches from text; the melody variant also follows an audio melody.
GenerationOpen-source code
- Start here
- Sketch an instrumental idea, or compare text-only generation with melody conditioning. Keep the non-commercial weight restriction in view.
- Training data
- 20,000 hours from Meta Music Initiative Sound Collection, Shutterstock and Pond5, with vocals removed for the public releases.
- Training rights
- The authors report agreements with rightsholders. These private collections are not redistributed with the weights.
- License & output terms
- Code: MIT. Weights: CC BY-NC 4.0.
- Limitations
- Non-commercial weights; weak realistic vocals, English prompt bias and uneven coverage of musical cultures.
- Technique
- Autoregressive audio tokens
- Runtime
- Local inference; memory depends on the small, medium, large or melody checkpoint.
- Parameters
- Not itemized for this release/family. Parameter sizes shown elsewhere may exclude encoders, decoders or other runtime components.
Sources reviewed . Training and rights notes summarize provider disclosures. A model license does not clear every source recording or generated output. Parameter counts refer to the named variant or component, not total runtime memory.