magenta

magenta-realtime: Generate Music Live from Text Prompts

Magenta RealTime 2 is an open-weights music generation model with Python and C++ inference libraries, aimed at live performance and app integration. Real-time streaming requires Apple Silicon; Python also supports offline inference on Apple Silicon or NVIDIA GPUs.

View on GitHub Compare
Stars
1.8k
Forks
201
Stars, last 30 days
+37
Last commit
2 months ago
Contributors
11

Our take

Activerecent releases

Health score

39/100

How it is scored
Activity
44
Community
53
Issues
21
Pull requests
36

+37 stars in the last 30 days

This is a focused, actively maintained toolkit for open-weights music generation, with a practical path from experimentation to Apple Silicon apps, but real-time deployment depends on compatible hardware.

Good fit if

  • You want to prototype or ship prompt-driven music generation and can use Python or C++.
  • Your real-time target is Apple Silicon, or your use case can run offline on Apple Silicon or an NVIDIA GPU.
  • You value an Apache-2.0 licensed project with recent commits and releases, and a contributor base that is not concentrated in one person.

Look elsewhere if

  • You need real-time streaming on hardware other than Apple Silicon.
  • Your project requires supervised fine-tuning today, which the repository describes as a future capability.
  • You need a project with recent issue closures as evidence of issue resolution; none were recorded in the measured 90-day period, despite merged pull requests.
All health signals
Last commit2026-07-30 (2 months ago)
Commits, last 90 days5
Releases, last 12 months1 (latest v2.0.3, 2026-07-30)
Contributors11 (top contributor: 48% of commits)
Issues closed, last 90 days0
Pull requests merged, last 90 days5 (typically merged in 2 days)
Project age16 months

Checked on 2026-10-08 with the GitHub API.

Project history

Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.

Selected snapshot

Stars

1,870

Forks

212

Change since Nov 15, 2026: +2 stars·0 forks

Overview

Magenta RealTime 2 generates music in response to prompts and is designed for streaming audio as it plays. The repository combines the model with Python inference backends and a C++ engine, so developers can experiment with generation or embed it in music software.

It is suited to musicians, researchers, and developers building interactive music tools. Real-time use is specifically oriented toward Apple Silicon, while the Python library also supports offline inference on Apple Silicon and NVIDIA GPUs.

Key Features

  • Open-weights model available in small and base sizes.
  • Python inference library with JAX and MLX backends.
  • C++ inference engine for streaming audio generation on Apple Silicon.
  • CLI commands for setting up models and generating audio.
  • Example applications include a DAW plugin, a standalone macOS app, and tools for exploring note control and prompt space.
  • Documentation covers installation, models, inference, exporting, benchmarks, and testing.

Use Cases

  • A musician or producer can try prompt-based live generation in a DAW using the AUv3 example plugin.
  • A developer building an interactive music app can embed the C++ engine on Apple Silicon or use the Python library for inference.
  • A researcher can experiment with model inference and prompts through the Python API and notebooks.
  • A performer using an Apple Silicon Mac can generate accompaniment or musical material in real time, choosing the small model for broader device support.

What you need

Detected in the repository

  • Python >=3.11 (from pyproject.toml)
  • A test suite and automated checks on GitHub Actions

License in plain words

Apache-2.0permissive

  • Commercial use: yes
  • Modify and redistribute: yes
  • You must keep: the license, the NOTICE file and a note of your changes
  • Share your changes: no
  • Includes an explicit patent grant from the contributors:

A summary, not legal advice: the LICENSE file is what applies.

Getting Started

On Apple Silicon, install the MLX extra and initialize the model resources:

uv pip install "magenta-rt[mlx]"
mrt models init
mrt models download
mrt mlx generate --prompt "disco funk" --duration 4.0 --model=mrt2_base

See the README for development setup and the full documentation for model and hardware details.

Alternatives

  • riffusion-hobby: Riffusion-hobby generates music from prompts with Stable Diffusion, while Magenta RealTime focuses on streaming generation and app integration.
ProjectLanguageLicenseStarsStatus
magenta-realtimePythonApache-2.01.8kActive
riffusion-hobbyPythonMIT3.9kInactive

Considerations

Real-time streaming is limited to Apple Silicon, and the larger model needs a high-end Pro Max chip for real-time operation. Python offline inference is also available on NVIDIA GPUs, but that does not make the C++ streaming path cross-platform. The project has recent releases and commits and merges pull requests, though the measured period showed no closed issues. It has not yet added the supervised fine-tuning support mentioned as a future update. The Apache-2.0 license permits broad use, subject to its license terms.

Found this useful?

Share it with someone who would like magenta-realtime.

Comparisons