AudioCraft

What it does

AudioCraft is a PyTorch code base from Meta AI for audio generation research. It takes text prompts (and, for some models, melody, chord, drum or style conditioning) and produces audio: music from MusicGen, environmental sound effects from AudioGen. It also includes the EnCodec neural codec, which compresses raw audio into discrete tokens and decodes them back to a waveform.

People use it by installing the Python package, running inference with the released models, or training their own models with the included pipelines. What sets it apart is that generation, compression and training live in one repository. The listed models include MusicGen, AudioGen, EnCodec, Multi Band Diffusion, MAGNeT, AudioSeal watermarking, MusicGen Style and JASCO.

It suits machine-learning researchers and developers comfortable with Python who want to experiment with, reproduce or build on audio generation models. Setup requires Python 3.9 and PyTorch 2.1.0, with ffmpeg recommended. It is a code library rather than a polished app for musicians. The code is MIT-licensed, but the model weights are released under CC-BY-NC 4.0, which restricts commercial use.

What it can do

• Text-to-music generation with MusicGen
• Text-to-sound-effect generation with AudioGen
• EnCodec neural audio compression and tokenization
• Training code for EnCodec, MusicGen, Multi Band Diffusion and JASCO
• Audio watermarking with AudioSeal
• Melody, style, chord and drum conditioning via MusicGen Style and JASCO

Example workflows

Generate a music clip from a text prompt

Input: a text description of the desired music. Steps: install audiocraft with pip, load a MusicGen model, run inference with the prompt. Output: a generated music audio sample.

Train a model on your own audio

Input: a dataset of audio files. Steps: clone the repository, install with pip install -e ., follow the training documentation and model-specific configs. Output: a trained model such as EnCodec or MusicGen.

Before you choose

• Model weights are CC-BY-NC 4.0 (non-commercial), while the code is MIT
• It is a research code base, not a hosted app with a user interface
• Training requires a local clone of the repository

Listed under

• Creative
• Music and audio
• Audio compression
• Developers
• Music generation
• Research
• Sound effects
• Text-to-audio

Get started

Cloning is required for training

1. Install PyTorch 2.1.0 on Python 3.9 first
2. Run python -m pip install -U audiocraft (or clone and pip install -e . for training)
3. Install ffmpeg and optionally set AUDIOCRAFT_CACHE_DIR for model downloads

Code is MIT-licensed and weights CC-BY-NC 4.0; no fees are listed. Compute and hardware costs are the user's.

AudioCraft is free to use. See their pricing page ↗

Sources for this overview

Product pages, documentation, and pricing checked 28 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Home ↗audiocraft.metademolab.com
• GitHub - facebookresearch/audiocraft: Audiocraft is a library for audio processing and generation with deep learning. It features the state-of-the-art EnCodec audio compressor / tokenizer, along with MusicGen, a simple and controllable music generation LM with textual and melodic conditioning. · GitHub ↗github.com

Company website

audiocraft.metademolab.com: product, pricing and policy pages

YC directory mirror

Batch and company name

Web search

Confirms the official site

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 28 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.