Open-source · Google Magenta · Apache 2.0

Generate MIDI inside Ableton. Locally. Free.

Five open-source AI tools — Continue, Generate, Groove, Drumify, Interpolate — that turn rigid MIDI into musical ideas. All models run on your machine. No cloud, no rate limits, no subscription.

FREE Max for Live RNN + VAE Local inference

From a single clip to a generated phrase in three steps.

Drag the Max for Live device onto any MIDI track in Ableton Live, point it at an existing clip, hit Generate. The model runs locally — no internet required.

01 — INSTALL

Drop the .amxd file onto a MIDI track

Download the bundle from the Magenta site, drag the Max for Live device onto an empty MIDI track in Ableton Live 10.1 Suite or higher. Five tools appear as buttons inside the device. No login, no API key, no extra installer.

requires Max for Live
02 — POINT & PICK

Select a clip in Session View, pick a tool

Click into a clip with some MIDI in it — a melody, a drum loop, a chord progression. Choose Continue, Generate, Groove, Drumify or Interpolate. Each tool uses a different ML model under the hood.

RNN · MusicVAE · GrooVAE
03 — GENERATE & KEEP

Hit Generate. Keep what works.

Output appears as a new MIDI clip in Session View. Tune the temperature slider for more or less randomness. Re-generate as many times as you want — local inference means no quotas. Keep what works, throw away the rest.

output · type-1 MIDI · stays in your project

Not one model. Five focused tools, each tuned for a specific MIDI job.

Most AI music tools wrap everything behind a single "Generate" button and hope. Magenta Studio splits the problem into five purpose-built modules — each backed by a different research model from Google Magenta's open library.

You don't need to know the difference between an RNN and a Variational Autoencoder to use them. But once you do, you'll pick the right tool for the job: Continue for organic extensions, Generate for fresh ideas, Groove to humanize quantized parts, Drumify to invent drums from a melody, Interpolate to morph between two clips.

The whole thing is open-source under Apache 2.0. Source code, training notebooks and model checkpoints live on the Magenta GitHub — researchers fork and retrain on their own datasets.

Built for people who'd rather edit MIDI than describe songs.

Magenta Studio isn't a prompt-to-song generator. It's a MIDI-level idea generator for producers who already think in clips, notes and patterns.

a

Ableton producers

You live in Session View. Magenta Studio drops in as a Max for Live device and starts generating ideas without leaving your set.

b

Electronic musicians

Get past writer's block on drum patterns and melodic phrases. Hit Generate with different temperature settings until something sparks.

c

ML researchers

Apache 2.0 source on GitHub. Train models on your own datasets, fork the device, contribute back to the open Magenta project.

d

Music educators

Teach machine learning through music — students see how RNNs continue a phrase, how VAEs interpolate between two ideas, all hands-on.

Open research, shipped as a plugin.

Magenta Studio is the production-friendly wrapper around years of Google's published music-ML research. Same models, friendlier UI.

Core

Every model runs locally on your machine.

No API keys, no usage caps, no cloud round-trips, no internet required. The trained model weights ship with the plugin and run on your CPU through TensorFlow.js. Generate as many times as you want — the only limit is how patient you are while the network thinks.

Pricing

Free forever, no signup

Apache 2.0 open-source. Download the .amxd, drag it in, generate. That's it.

Temperature

Tune randomness from safe to weird

One slider per tool controls how adventurous the output is. 0.5 = predictable, 1.5 = surprising.

Session View

Generated clips drop into your set

Output appears as new MIDI clips inside Ableton — auditioning, editing and arrangement stay native.

Open research

Source code & weights on GitHub

Forkable, retrainable, modifiable. Research papers published alongside every model release.

Privacy

Your MIDI never leaves your machine.

Unlike SaaS music tools that send your clips to a cloud endpoint for inference, Magenta Studio loads the model weights into your local Ableton process. Useful for confidential work, sketches under NDA, or anyone who'd just rather not upload their ideas.

What Magenta is built for — and what it isn't.

Different problem space than Suno or Splice. Magenta is a MIDI-level idea engine that lives inside Ableton — nothing else in the market quite competes with it directly.

Magenta Studio Ableton native Suno Splice
PriceFreew/ Suite$10/mo$10/mo
AI MIDI generationYesNoNo (audio only)No
Runs locally (no cloud)YesYesNoNo
Works in Ableton LiveNativeNativeNoPlugin
Works in FL / Logic / ReaperNoNoNoYes
Open-sourceApache 2.0NoNoNo
Full-song generationNoNoBest-in-classNo
Pricing & features verified June 2026 from publisher pages.

What producers say after living with the device.

★★★★★
I've been using Continue and Drumify for two years. The melodies it suggests are not always great, but they unstick me. Hit Generate three times, one of them sparks something, I edit from there. The fact that it's free, local, and I can run it on a plane is what made it stick.
F
Finn O.
Electronic producer, Dublin
★★★
Honest take: the UI is ugly, the docs are sparse, the project hasn't had a major update in a while, and it's Ableton-only. But the underlying models are good and the temperature slider is genuinely useful. Best as a research-y idea machine, not a polished product.
R
Riya P.
Beatmaker, Bangalore
★★★★★
I teach a music-ML class at a community college. Magenta Studio is the only tool I've found where students can hear an RNN extend a phrase or a VAE interpolate between clips in real time, in a DAW they actually know. The fact that the source is on GitHub means they can read the code afterwards.
N
Nadia E.
Music tech educator, Toronto

A research project, packaged as a plugin.

Magenta started inside Google Brain in 2016 as an open research effort exploring whether machine learning could generate convincing music and art. The team published papers, released model checkpoints, and built tools around them — but most of it lived in Python notebooks and command-line scripts, far from any practising musician's workflow.

Magenta Studio shipped in late 2018, first shown publicly at Ableton's Loop conference in Los Angeles. The idea was simple: take the most production-friendly Magenta models — Continue, Generate, Groove, Drumify, Interpolate — and wrap them in a Max for Live device so that anyone with Ableton Live Suite could try them without writing a line of code. The source went up on GitHub under Apache 2.0.

It's important to be honest about the trade-offs. The plugin is Ableton-only — there's no FL Studio, no Logic, no Reaper, no Pro Tools build. The UI looks like research-team UI: it works, but it's not winning design awards. Updates ship at research pace rather than commercial pace; there are stretches where the GitHub repo goes quiet for months. And because the models are trained on relatively old datasets, the generated phrases sometimes have a recognizably 2018-era flavour rather than the latest aesthetic.

If you don't use Ableton, this isn't the tool — pick a DAW-agnostic plugin. If you want one-prompt finished songs, this isn't that either — go to Suno. But if you're an Ableton producer who'd like a hands-on MIDI-level AI that runs locally, is open-source, and costs nothing, Magenta Studio is the most credible thing in that exact category.

Questions, answered.

Is Magenta Studio actually free, or is there a paid tier?

Genuinely free — Apache 2.0 open-source, no paid tier, no subscription, no usage caps. Download the plugin bundle, drop the Max for Live device onto a MIDI track, generate as much as you want. The only "cost" is owning Ableton Live 10.1 Suite (or higher) with Max for Live, which is a paid Ableton product — but Magenta itself charges nothing on top of that.

Does it work in FL Studio, Logic Pro, or Reaper?

No — Magenta Studio is built as a Max for Live device and only loads inside Ableton Live. A standalone version exists in older releases but isn't actively maintained and lacks the workflow integration that makes the plugin useful. If you don't use Ableton, this isn't the right tool; pick a DAW-agnostic AI MIDI plugin instead.

What's the difference between Continue and Generate?

Continue takes an existing MIDI clip and extends it — the model listens to what's there and predicts a musical continuation, up to 32 measures. Generate ignores your input and creates a new clip from scratch, either drums or a monophonic melody. Continue is for "what comes next?" Generate is for "give me a starting idea."

What does the temperature slider actually do?

Temperature controls randomness in the model's sampling. Low temperature (0.5) makes the output predictable and close to common patterns — safer, less surprising. High temperature (1.2–1.5) lets the model explore unlikelier options — more interesting, more chaotic, more likely to produce something unusable. Most producers settle around 0.9–1.1 for usable variation.

Does it require an internet connection?

No. Once you've downloaded the plugin, model weights live inside the Max for Live device and run on your CPU. You can generate on a plane, in a basement studio, behind a corporate firewall — anywhere Ableton runs. Your MIDI clips never leave your machine.

How musically convincing is the output?

Honest answer: variable. The drum tools (Groove, Drumify) and the Continue tool on short phrases produce material that often goes straight into a project. Melodic generation is more hit-or-miss — useful as an idea-spark, less so as a finished melody. Most producers treat the output as a starting point: hit Generate three or four times, pick the seed that sparks something, then edit by hand.

Is the project still active?

Yes, but slowly. Magenta moves at research pace, not commercial-product pace. The GitHub repos receive periodic updates, papers are still published, but you won't see monthly feature releases like a paid SaaS. Major versions of the plugin are years apart. The current plugin is stable and works on modern Ableton versions even when updates are infrequent.

Can I train my own model and use it inside the plugin?

Yes, with some engineering work. The plugin loads pre-trained checkpoints, but the entire training pipeline is published on the Magenta GitHub. Researchers and producers with ML experience have trained variants on jazz datasets, classical corpora and personal sample libraries, then loaded the resulting weights into a forked version of the device. Documentation is limited — expect to read code.

Stop staring at the empty bar. Hit Generate.

Free, open-source, runs locally inside Ableton. Five tools, infinite ideas.

Magenta Studio
Free · open-source · Ableton plugin
Download