Skip to content
EN
English 简体中文 soon 日本語 soon

AudioCraft

Meta AI generative audio library

Visit official site

What AudioCraft is

AudioCraft is Meta AI's generative audio library rather than a consumer product. It brings three pieces together: MusicGen for music from text, AudioGen for sound effects from text, and EnCodec for neural audio compression. Both generation models come with browser demos, so anyone can try them without setting anything up.

What you can do with it

  • Generate musical passages from a prompt
  • Create environmental ambience and effects
  • Control described rhythm, atmosphere and duration through prompting
  • Study how the underlying models behave
  • Prototype audio ideas before building your own pipeline

Who it is for

  • Students and researchers studying generative audio
  • Sound designers experimenting with prompts
  • Developers comfortable running the code locally

What to watch out for

  • As a research release, licence terms restrict commercial use in ways you must read before distribution
  • Output length and fidelity lag commercial generators
  • Running the models yourself needs capable hardware
  • Training data provenance is a live question in music generation; check what the models were built on

Pros & cons

✓ What we like

  • Open research code with published demos
  • Covers both music and sound effects
  • Good starting point for learning

! What to watch out for

  • Research licence restricts commercial use
  • Requires hardware to run locally
  • Quality below commercial generators

FAQ

Is it a product or a library?

It is a research library with code and online demos published by Meta AI.

What can it generate?

Music through MusicGen and sound effects through AudioGen, with EnCodec handling compression.

Can I ship what it produces?

Check the licence carefully, since research releases often exclude commercial use.

Last reviewed: 2026-09-18

More AI music creation tools

View all →

How we review