Meta AudioCraft (MusicGen)

AudioCraft is Meta’s open-source audio research library containing MusicGen, AudioGen, and EnCodec for text-to-music and sound generation.

Overview

AudioCraft bundles Meta’s generative audio research into an open-source Python library. MusicGen turns text descriptions or melodies into instrumental music, AudioGen generates sound effects and environmental audio, and EnCodec provides neural audio compression. All models ship with weights, enabling researchers and developers to run, study, and fine-tune them locally.

As an open project under permissive licenses for research and many commercial uses, AudioCraft has become the foundation for countless music-generation experiments and products. Pretrained models range from 300M to 3.3B parameters, balancing quality and hardware requirements for different use cases.

For builders, AudioCraft’s real value is control: weights, training code, and inference scripts are all available, so teams can fine-tune on proprietary sound libraries or distill smaller models for on-device use. The research papers accompanying each release explain the architecture decisions clearly.

Key Features

• MusicGen text-to-music generation with melody conditioning
• AudioGen text-to-sound-effects and environmental audio
• EnCodec neural audio codec for high-quality compression
• Open weights and code for self-hosting and research
• Multiple model sizes from 300M to 3.3B parameters
• Melody conditioning to guide generation with reference audio
• Active research community building on the library
• Full training code access for fine-tuning on proprietary audio
• Model distillation possibilities for on-device deployment

Pricing

AudioCraft is completely free and open source. Users can download models and run them on their own hardware without licensing fees, subject to Meta’s model license terms for each component. Self-hosting costs only compute, which makes AudioCraft the most economical option for organizations with GPU capacity and engineering talent.

Who Is It For

AudioCraft is built for AI researchers, developers, and hobbyists who want to experiment with music generation models without vendor lock-in.

It suits teams embedding audio generation into their own products.

Startups building music features often start from AudioCraft to validate product ideas before investing in proprietary model training.

FAQ

Is AudioCraft free to use?

Yes — the code and model weights are open source and free to use under Meta’s licenses.

Does AudioCraft offer an API?

No hosted API is provided; users run the models themselves or use third-party wrappers and services.

How does MusicGen compare to Suno or Udio?

MusicGen produces instrumental music and is fully open for self-hosting, while Suno and Udio generate complete songs with vocals through hosted services.

What hardware do I need to run MusicGen?

The smaller model variants run on consumer GPUs with reasonable speed, while the largest 3.3B-parameter model benefits from high-memory data-center GPUs for comfortable generation times.