MelodyFlow

MelodyFlow

Runs a demo of audio generation models from the facebook group, built from the AudioCraft library. For people who want to try music and sound generation in a browser or read the code.

Made by facebook on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: MelodyFlow is a Gradio demo from the facebook group, and its repository is the AudioCraft library, a PyTorch toolkit for audio generation research. The demo starts from the file demos/melodyflow_app.py and names several models, among them MAGNeT, AudioGen, AudioSeal and two EnCodec versions. The card describes the library more than the demo itself.

Who it suits: Audio researchers and developers who want to try generative audio models in a browser or read how the code is organised. Curious beginners can also open the Space and see what the demo offers.

Industry | Who uses it | How they use it | Example
Research | Audio and machine learning researchers | Run the demo and read the AudioCraft code behind it | Open the Space, then look through the audiocraft folder in a local copy
Music and media | Sound designers | Try text-driven music and sound models | Test a short text description in the demo
Software | Developers | Check how the Gradio demo is wired to the models | Read demos/melodyflow_app.py in a copy

If you want a ready product for commercial use, you can skip it, because the model weights carry a non-commercial license.

Where to use it: The demo runs in these places:

• A web browser: open the Space page on Hugging Face and use it there.
• Windows, macOS or Linux: run a copy on your own computer with Python 3.10.13.

Use case | Who | What they get back
Try the demo online | Anyone curious about generated audio | Audio from the page's controls
Study the library | Researchers | Training and inference code for audio models, with docs and tests
Run it locally | Developers | The same Gradio demo on their own computer

How to set it up: You need a few things first:

• Python 3.10.13, the version this Space is set to
• The huggingface_hub package, which the first command installs
• The packages listed in requirements.txt
• ffmpeg, which the AudioCraft README recommends having installed

The steps download the Space, which holds the whole AudioCraft repository, install Gradio at the version the Space uses, add the other packages and start the demo file.

1. Install the Hugging Face package.
2. Download the Space into a folder called MelodyFlow.
3. Move into that folder.
4. Install Gradio 5.49.1.
5. Install the packages from requirements.txt.
6. Start the demo file.

pip install -U huggingface_hub
hf download facebook/MelodyFlow --repo-type space --local-dir MelodyFlow
cd MelodyFlow
pip install gradio==5.49.1
pip install -r requirements.txt
python demos/melodyflow_app.py

The README suggests ffmpeg through your system on Linux:

sudo apt-get install ffmpeg

How to use it: Open the Space page on Hugging Face, or start the local copy with the last demo command above. Gradio serves a web page for the demo. The README also describes the library's model families.

python demos/melodyflow_app.py

Tool | What it does
MusicGen | A controllable text-to-music model
AudioGen | A text-to-sound model
EnCodec | A high fidelity neural audio codec
Multi Band Diffusion | A diffusion decoder that works with EnCodec
MAGNeT | Makes music or sound from a text prompt without generating one step at a time
AudioSeal | An audio watermarking model

Example: You run python demos/melodyflow_app.py and open the page it serves. The README lists text-to-music and text-to-sound models among the code, so a short written description of a sound is a natural first input, and you would expect audio back.

For newcomers and experts: If you are new to this, open the Space page and try the demo once. You do not need to install anything for that. A local copy is a big download, since the repository has 287 files, so try the online page first.

If you know the field, the repository is AudioCraft, with inference and training code, configs, docs and tests. The README says the code is MIT and the weights are CC-BY-NC 4.0. It names Python 3.9 and PyTorch 2.1.0 for the library, while this Space is set to Python 3.10.13.

You open the Space or run the demo file
A Gradio page takes your input
AudioCraft models work on it
The result shows on the page

Good to know: The card does not describe MelodyFlow's own controls, inputs or output length.

• License: cc-by-nc-4 for the Space. The README says the code is MIT and the model weights are CC-BY-NC 4.0.
• Cost to run: none named. The hosted Space runs on Hugging Face hardware of the zero-a10g type.
• Data it reaches: what you give the demo, plus model files stored in a cache folder that the AUDIOCRAFT_CACHE_DIR setting can move.
• Not for: commercial use of the model weights, since their license is non-commercial.

On Hugging Face, 13 November 2026

• Likes: 168
• Built with: gradio
• License: cc-by-nc-4.0

Links

• Space: https://huggingface.co/spaces/facebook/MelodyFlow
• Files: https://huggingface.co/spaces/facebook/MelodyFlow/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.