Meta's open music generation model, runnable locally or on Hugging Face
MusicGen is Meta AI's open-source music generation model, published with weights, code and a demo Space on Hugging Face. You describe the music you want - genre, instruments, mood, tempo - and can optionally supply a reference melody, and the model returns a short instrumental clip. Because the weights are open, it can run on your own GPU or inside another application rather than through a paid service. Quality is that of a research model: useful for sketches, loops and sound beds, and a common baseline for anyone building on music generation.
Last updated: 2026-09-20. This site only provides an index; for exact features, pricing, and licensing, see the official website.
music sketches, game loops, research experiments and audio prototyping
If you're comparing similar products, check the alternatives below, or browse all tools in the AI Audio & Music category.
The code and weights are released for broad use, but licensing terms and the training data are separate questions addressed in the project documentation. Review the model card before shipping anything commercial.
No. It produces instrumental audio. Prompting for singing generally yields indistinct vocal-like textures rather than clear words, so pair it with a separate voice tool if you need lyrics.
Individual generations are short by design, and longer pieces are built by chaining outputs. That works for loops and beds, but stitched transitions need listening to and often a little editing.
Text-to-speech voices that are hard to distinguish from humans
Professional audio workstation for editing, restoration and podcast mixing
Browser audio cleanup that makes phone recordings sound studio grade
Automatic audio post-production for loudness, levels and noise