Audio Gen · Model Guide · 4 min read

MusicGen Medium 1.5B

Meta's open music generator — instrumental loops and melodies from text, running locally on a mid-range GPU.

Basic Information

MusicGen is Meta's open music generation family: text-prompted instrumental music in 30-second segments, with melody-conditioning so you can hum or feed a reference melody. The medium model (~1.5B parameters) balances quality and hardware cost; the small model runs even lighter, and the large one trades speed for detail.

It is instrumental-only — no sung vocals like Suno — but for local, license-clear background music and sound palettes it remains the standard open option, with melody conditioning that lets you hum an idea and have it orchestrated.

Best use cases

• Local, royalty-question-free instrumental beds for videos and games (mind the license)
• Melody-conditioned composition: hum a tune, get an arrangement
• Batch sound-palette generation for apps and installations
• Offline pipelines where hosted music services are not an option

Computational Requirements

Modest: the medium model runs on 8–16 GB GPUs; generation of 30-second clips takes well under a minute.

Commercial license

Weights are released under CC-BY-NC 4.0 — non-commercial use only. This is the important caveat: unlike most models on this page, you cannot legally use MusicGen outputs in commercial products; use it for prototyping and personal projects, and mind that Meta's model licenses differ from their Llama line.