Stable Audio Open
Open-weight text-to-audio — 47-second sound effects and music.
Meta's text-to-music & sound-effect model family.
Runs locally · Entry GPU (6–8 GB)
MusicGen small/medium fits in 8 GB; large needs 16 GB.
Source is public — you can audit it, fork it, and you'll never lose access to your workflows if AudioCraft (MusicGen) the company changes direction.
Comfortable on a mid-range consumer card — no need to remortgage for an A100.
Native Metal / MPS support — runs on M-series Macs without CUDA gymnastics.
You don't need to read a paper before getting your first result — sensible defaults and a quick install.
This page summarizes upstream documentation, release information, and editorially reviewed catalogue fields. It is not presented as a hands-on benchmark. Verify changing requirements at the official project; report stale data through our corrections channel.
A standard local install — download, install dependencies, point at your GPU.
Needs 8 GB minimum — RTX 3060 12GB or 4070 territory.
Source is public — auditable and forkable, no vendor lock.
727 days since the last refresh — treat hardware numbers as a floor, not a ceiling.
Three picks across different tradeoffs — so you don't end up with three near-clones of AudioCraft (MusicGen).
AudioCraft bundles MusicGen, AudioGen, and EnCodec from Meta FAIR. MusicGen produces 30-second music clips from a text prompt or a melody reference; AudioGen does ambient sound effects. The benchmark OSS baseline for music generation before Stable Audio Open.
MIT-licensed code; CC-BY-NC weights (non-commercial).
Open-weight text-to-audio — 47-second sound effects and music.