Magenta RealTime 2
Google's open live-music model: it plays continuously on a MacBook and reacts to text, audio and MIDI in about 200 ms.
In short. Magenta RealTime 2 is less a song generator than a live instrument: it keeps making music as you play, and you steer it with text, a sound example or MIDI notes. It runs on an Apple Silicon MacBook, and the weights are CC BY 4.0.
Summary
Google DeepMind's Magenta team published Magenta RealTime 2 (MRT2) on 4 June 2026, as a follow-up to the original Magenta RealTime and the Lyria RealTime API. Instead of turning a prompt into a finished track, it generates 48 kHz stereo audio continuously, one 40-millisecond frame at a time, and takes control signals at every frame: a style given as text or an audio clip, plus MIDI notes and drum on and off signals. Google says control latency is about 200 ms, roughly 15 times lower than the first version, and that at release it was the only open-weights model for real-time, continuous music generation with low-latency control.
The model is a decoder-only transformer over SpectroStream codec tokens, with MusicCoCa embeddings for style, in two sizes: Base with 2.4 billion parameters and Small with 230 million. Google's blog says Base needs a MacBook with an M3 Pro or M2 Max or better, while Small runs on any Apple Silicon MacBook, including the Air. The release includes a Python library for JAX and MLX, a C++ inference engine built on MLX, a plugin bundle for macOS and example apps that run it standalone or inside a digital audio workstation. It was trained on about 71,000 hours of stock music, mostly instrumental, so it makes instrumental music with only non-lexical vocal sounds. Google lists fine-tuning support as coming.
The code is Apache 2.0 and the weights are CC BY 4.0. Google adds terms that ask you to use the materials responsibly and not to generate output that infringes others' rights, and it claims no rights in what you generate. In the 30 days to 11 November 2026, Hugging Face counted about 19,600 downloads.
Good for
- Live performance, improvisation and interactive installations controlled by MIDI, text or audio
- Game and app soundtracks that react to the player in real time
- Instruments and plugins inside a digital audio workstation on a Mac
- Research on streaming music generation, since the code and weights are open
Watch out for
- It makes instrumental music rather than songs with lyrics, and its vocal sounds are non-lexical
- The Base model asks for an Apple Silicon MacBook with an M3 Pro or M2 Max or better, and Google's real-time engine and apps are built for Macs
- CC BY 4.0 requires credit to Google, and Google's terms forbid infringing output
- Fine-tuning support is listed as coming, so you cannot customize the model yet
Performance claims come from the developer and have not been independently verified.
How much memory does it need?
Memory figures for each variant of Magenta RealTime 2. Image, video, speech and music pipelines also load a text encoder and a decoder, so the memory needed can be larger than the main checkpoint.
| Variants | Parameters | Download size | Memory needed | Fits on |
|---|---|---|---|---|
| Small (230M) FP32 | 230M | 2.2 GB | 2.53 GB | Laptop or phone |
| Base (2.4B) FP32 | 2.4B | 11 GB | 13 GB | Laptop or phone |
Download size covers every component needed to run the model, such as the generator, text encoder and decoder. Memory needed is a stated or published measured minimum where one exists; otherwise it is the download size plus about 15%. Larger images, longer clips and higher resolutions need more.
The license in plain words
A Creative Commons license that lets anyone copy, share and adapt the work for any purpose, including commercial use, as long as they give credit. It was written for creative and data works rather than software, but some speech and data releases use it for model weights.
- Attribution or model name must be shown
See also
- Best open-source music and sound-generation models in 2026
- Open music models grew up in 2026: seven models for songs, instruments and sound effects
Frequently asked questions
- What is Magenta RealTime 2?
- Magenta RealTime 2 is less a song generator than a live instrument: it keeps making music as you play, and you steer it with text, a sound example or MIDI notes. It runs on an Apple Silicon MacBook, and the weights are CC BY 4.0.
- Can I use Magenta RealTime 2 commercially?
- Yes. Magenta RealTime 2 is released under CC BY 4.0, a permissive license. You can use it commercially if you follow the standard conditions, such as keeping the license notice.
- Is Magenta RealTime 2 open source?
- Magenta RealTime 2 is open-weight, not open source in the strict sense. You can download the model files, but the CC BY 4.0 adds conditions that the Open Source Initiative definition does not allow.
- Can I run Magenta RealTime 2 on my own computer?
- Yes. The smallest variant, Small (230M), needs about 2.53 GB of memory at 4-bit precision, which fits a modern laptop or a mid-range graphics card.
- When was Magenta RealTime 2 released?
- Magenta RealTime 2 was released by Google DeepMind on June 4, 2026.
Official links
Sources
Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.
- Magenta RealTime 2: Open & Local Live Music Models (Magenta team, 4 Jun 2026) magenta.withgoogle.com
- Magenta RealTime 2 model card, with terms of use (Hugging Face) huggingface.co
- Magenta RealTime repository (GitHub) github.com
- Live Music Models (the Magenta RealTime report, arXiv) arxiv.org
Release date and facts last verified November 11, 2026.
Related models
| Model | Released | Parameters | Context | License | Smallest variant needs |
|---|---|---|---|---|---|
| Stable Audio 3 Stability AI | 460M – 1.4B | – | Stability Community (Conditional) | 4.02 GB Laptop or phone | |
| ACE-Step 1.5 ACE-Step | 2B – 4B | – | MIT (Permissive) | 4 GB Laptop or phone | |
| MiniMax Music 3 MiniMax | 11.1B | – | MiniMax Music3 Community (Conditional) | 24 GB Gaming GPU | |
| MOSS-SoundEffect v2.0 OpenMOSS | 1.3B | – | Apache 2.0 (Permissive) | 13 GB Laptop or phone |