AudioCraft

AudioCraft is Meta’s open-source PyTorch library for deep learning research on audio generation — MusicGen, AudioGen and EnCodec.

Overview

AudioCraft packages Meta’s audio generation research into an open-source library: MusicGen generates music from text prompts, AudioGen generates sound effects, and EnCodec is a neural audio codec. Researchers and developers can run, fine-tune and build on these models at no cost, and the weights are published for both research and commercial use under Meta’s license. It is the reference open implementation behind a wave of music generation experiments.

Key Features

• MusicGen text-to-music generation models
• AudioGen text-to-sound-effect generation
• EnCodec neural audio compression
• Open-source code and model weights
• Fine-tuning support for custom datasets
• Research-friendly PyTorch implementation

Pricing

AudioCraft is free and open source; costs come only from the compute needed to run the models.

Who Is It For

Researchers, developers and tinkerers who want to run and customize music and audio generation models locally.

Use Cases

Researchers study text-to-music generation and fine-tune models on niche genres, and game developers prototype procedural audio with AudioGen. Hobbyists run MusicGen locally to create royalty-free background tracks.

Getting Started

Clone the repository, install dependencies and run the provided MusicGen notebook on a GPU machine for your first generations.

FAQ

Is AudioCraft free to use?

Yes — completely free under Meta’s open-source license.

Does AudioCraft offer an API?

It is a library rather than a service; you run it yourself or wrap it in your own API.

How does AudioCraft compare to alternatives?

Versus Suno or Udio, AudioCraft offers openness and customization at the cost of convenience and output polish.

What hardware do I need?

A CUDA GPU is recommended for reasonable generation speed; smaller models can run on consumer cards.

Can outputs be used commercially?

Review the license attached to each model and your outputs’ terms before commercial use.

How good is the output quality?

MusicGen produces surprisingly coherent instrumental tracks for an open model, though commercial products like Suno currently lead on polish and song structure.

Can I fine-tune on my own music?

Yes — training scripts support fine-tuning on custom datasets, which is how researchers adapt the models to specific genres and styles.

What models ship with AudioCraft?

MusicGen for text-to-music, AudioGen for sound effects and EnCodec for neural audio compression — a complete open toolkit for audio generation research.

Is commercial use allowed?

Licensing terms are attached per model and code; review the repository licenses carefully before shipping commercial products.