MusicGen

Spec sheet

Purpose

MusicGen is a model for controllable music generation, intended primarily for AI research.Read 12 Oct 2026 · github.com

Supported platforms

webRead 21 Sep 2026 · audiocraft.metademolab.com

API

The documentation says AudioCraft provides a simple API and pretrained MusicGen models.Read 12 Oct 2026 · github.com

Support

The model card directs questions and comments to the GitHub repository or its issue tracker.Read 12 Oct 2026 · github.com

License

The repository identifies the code as MIT-licensed and the model weights as CC-BY-NC 4.0 licensed.Read 12 Oct 2026 · github.com

API and framework

The repository provides a simple API and documents MusicGen use through the Hugging Face Transformers library.Read 12 Oct 2026 · github.com

Audio export formats

wavRead 21 Sep 2026 · audiocraft.metademolab.com

Audio quality

The page says a Multi-Band Diffusion EnCodec decoder can reduce audio artifacts at the cost of more computation.Read 12 Oct 2026 · audiocraft.metademolab.com

Audio upload

YesRead 21 Sep 2026 · audiocraft.metademolab.com

Commercial use

NoRead 21 Sep 2026 · audiocraft.metademolab.com

Compute requirement

The documentation says inference with medium-sized models requires a GPU with at least 16 GB of memory.Read 12 Oct 2026 · github.com

Hardware requirement

Running inference with medium-sized models requires a GPU with at least 16 GB of memory, while smaller GPUs can generate short sequences or use the small model for longer sequences.Read 12 Oct 2026 · github.com

Known limitations

The model card says MusicGen cannot generate realistic vocals, performs less well with non-English descriptions, and does not perform equally well across music styles and cultures.Read 12 Oct 2026 · github.com

Licensing

The code is released under MIT, while the model weights are released under CC-BY-NC 4.0.Read 12 Oct 2026 · github.com

Long samples

The page describes generating two-minute tracks by sliding 30-second windows in 10-second chunks and retaining the previous 20 seconds as context.Read 12 Oct 2026 · audiocraft.metademolab.com

Longer tracks

The maker describes generating tracks up to two minutes by sliding 30-second windows in 10-second chunks while retaining 20 seconds of prior output as context.Read 12 Oct 2026 · audiocraft.metademolab.com

Melody guidance

MusicGen can extract a melody from an input music sample using chromagrams and guide generation to follow it while adhering to a text description.Read 12 Oct 2026 · audiocraft.metademolab.com

Model architecture

MusicGen uses a single-stage Transformer model with an EnCodec tokenizer.Read 12 Oct 2026 · github.com

Model design

MusicGen is a single-stage Transformer language model that generates music using compressed audio tokens.Read 12 Oct 2026 · github.com

Model options

The repository documents small, medium, and large models, as well as melody-conditioned and stereo variants.Read 12 Oct 2026 · github.com

Model sizes

The model card lists 300M, 1.5B, and 3.3B parameter sizes, with text-to-music and melody-guided variants.Read 12 Oct 2026 · github.com

Text generation

The model generates music samples from text descriptions.Read 12 Oct 2026 · audiocraft.metademolab.com

Text prompts

It generates music conditioned on textual descriptions.Read 12 Oct 2026 · audiocraft.metademolab.com

Training data

The documentation says MusicGen was trained on 20,000 hours of licensed music, including an internal set and music data from Shutterstock and Pond5.Read 12 Oct 2026 · github.com

Training-data availability

The repository says it does not provide the datasets used to train MusicGen and instead includes a small illustrative dummy dataset.Read 12 Oct 2026 · github.com

Vocal generation

NoRead 21 Sep 2026 · audiocraft.metademolab.com

Ways to use

The project documents a Hugging Face demo, a Colab notebook, a local Gradio demo, and a local Jupyter notebook.Read 12 Oct 2026 · github.com

Where Specifiction places it

• AI music generators#105 of 1237.3