AudioCraft
MetaAudioCraft is Meta's PyTorch library for audio generation research, carrying both inference and training code. It hosts MusicGen and MAGNeT for text-to-music, AudioGen for text-to-sound, JASCO for chord- and melody-conditioned music, the EnCodec neural codec, Multi Band Diffusion decoding and the AudioSeal watermark. The model weights it loads are published separately under their own terms.
Scored as the library. The MusicGen weights it loads are their own entry, and their non-commercial terms are recorded there.
Openness
5 high confidence- license
- MIT(OSI
- source
- public(facebookresearch/audiocraft)
- core features withheld
- no — no ee, enterprise, commercial or proprietary path
The AudioCraft code is MIT-licensed and fully public, training code included, and Meta offers no paid edition of it. The model weights the library loads are licensed separately for non-commercial use only, which is recorded on MusicGen rather than here.
- https://raw.githubusercontent.com/facebookresearch/audiocraft/main/LICENSE recorded 2026-09-26
LICENSE body: "MIT License / Copyright (c) Meta Platforms, Inc. and affiliates."
- https://raw.githubusercontent.com/facebookresearch/audiocraft/main/LICENSE_weights recorded 2026-09-26
LICENSE_weights body: "Attribution-NonCommercial 4.0 International", covering the model weights rather than the code.
- https://raw.githubusercontent.com/facebookresearch/audiocraft/main/README.md recorded 2026-09-26
README: "The code in this repository is released under the MIT license" and "The models weights in this repository are released under the CC-BY-NC 4.0 license"; no paid or hosted edition is named.
- https://ungh.cc/repos/facebookresearch/audiocraft/files/main recorded 2026-09-26
Full file list: no ee/, enterprise/, commercial/ or proprietary/ path; LICENSE and LICENSE_weights are the license files.
Adoption
1 high confidenceAdoption is measured as PyPI downloads of the audiocraft package, which is the library itself. Installs made directly from the repository are not counted.
- https://pypi.org/pypi/audiocraft/json recorded 2026-09-26
PyPI audiocraft: project_urls Homepage = https://github.com/facebookresearch/audiocraft; last release 1.3.0, uploaded 2024-06-03.
- https://pypistats.org/api/packages/audiocraft/recent recorded 2026-09-26
pypistats for audiocraft: last_month 9599.
Capability
2 high confidenceAudioCraft runs and trains several of Meta's music and sound-effect model families. It is a step below stable-diffusion.cpp because it covers a single modality, audio.
- https://raw.githubusercontent.com/facebookresearch/audiocraft/main/README.md recorded 2026-09-26
README Models: MusicGen ("text-to-music"), AudioGen ("text-to-sound"), EnCodec, Multi Band Diffusion, MAGNeT, AudioSeal, MusicGen Style and JASCO; "AudioCraft contains inference and training code".
Verified 2026-09-26