Multitrack Midi Music Generator

Multitrack Midi Music Generator

Generates multitrack MIDI music with a language model trained on 8-bar pieces. Runs in the browser as a Docker app, for musicians and developers who want to try AI-made MIDI.

Made by juancopi81 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: Multitrack Midi Music Generator makes music as MIDI, with several tracks at once. It uses two models named in the Space, a tokenizer and a language model whose names mention 8 bars. The Space runs on Hugging Face, so you can try it in a browser with nothing to install.

Who it suits: Musicians and hobbyists who want AI-made MIDI to play with, and developers who want to study a small Python music app packaged in Docker.

Industry | Who uses it | How they use it | Example
Music | Composers and hobbyists | Generate MIDI pieces as starting ideas | Create a piece and open it in a music editor
Games | Indie developers | Get quick musical sketches to try in a project | Generate a short piece and test it in a prototype
Machine learning | Researchers | Look at how a tokenizer and language model make music | Read model.py and the tokenizer files
Software | Developers | Build and run the Docker image locally | Run the container and publish port 7860

Anyone who needs finished audio recordings rather than MIDI can skip it.

Where to use it: It runs in these places.

• A web browser, on the Space's page at Hugging Face
• Windows, with Docker
• macOS, with Docker
• Linux, with Docker

Use case | Who | What they get back
Generate a multitrack piece | Musicians | A MIDI piece with several tracks
Collect ideas for a project | Game and video makers | Short musical sketches in MIDI form
Study the code | Developers | A local copy with main.py and model.py to read

How to set it up: Using the Space online needs only a browser. A copy on your own computer needs the following.

• Docker
• Python and pip, for the huggingface_hub package that gives you the hf command
• Disk space for the repository, which holds 11 files

1. Install the huggingface_hub package.
2. Download the Space's files into a local folder.
3. Move into that folder.
4. Build the Docker image and name it multitrack-midi-music-generator.
5. Run the container with port 7860 published to your computer.

pip install -U huggingface_hub
hf download juancopi81/multitrack-midi-music-generator --repo-type space --local-dir multitrack-midi-music-generator
cd multitrack-midi-music-generator
docker build -t multitrack-midi-music-generator .
docker run -it -p 7860:7860 multitrack-midi-music-generator

How to use it: Open the Space's page on Hugging Face, or start the container and open your browser on port 7860 of your computer. Then ask it to generate a piece and listen to or download the result.

docker run -it -p 7860:7860 multitrack-midi-music-generator

Tool | What it does
main.py | The main file of the app
model.py | Holds the code that works with the music model
string_to_notes.py | Converts the model's text output into notes
docker build | Builds the image from the Dockerfile
docker run | Starts the container and publishes port 7860

Example: You start the container, open the page on port 7860 and generate a piece. You get back a multitrack MIDI result that you can open in a MIDI editor or player.

For newcomers and experts: MIDI is a file of notes and timing, not recorded sound, so any music program can play it with the instrument you pick. The first small thing to try is the online Space, with one generated piece.

The model names point to a tokenizer for 8-bar pieces and a language model trained for 40 epochs. The model writes music as text tokens, and a helper turns them into notes. The repository splits this across constants.py, model.py, utils.py and string_to_notes.py.

You ask for a piece
The language model writes music tokens
string_to_notes.py turns them into notes
The page returns a multitrack MIDI piece

Good to know: The Space states no usage limits, so treat it as a demo you can read and adapt.

• License: mit
• Cost to run: none named
• Data it reaches: the Space files and the two model files that Hugging Face supplies
• Hardware: the Space is set to cpu-basic at Hugging Face, and what your own computer needs is not stated

On Hugging Face, 13 November 2026

• Likes: 130
• Built with: docker
• License: mit

Links

• Space: https://huggingface.co/spaces/juancopi81/multitrack-midi-music-generator
• Files: https://huggingface.co/spaces/juancopi81/multitrack-midi-music-generator/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.