Skip to main content
Music Maker AI
Open model · 2023

MusicGen AI music generator

Type a sentence and hear instrumental music start to play within the same run. A free community demo, ready below.

MusicGen is Meta’s open text-to-music model from 2023. It doesn’t produce realistic singing, and Meta released its trained weights for non-commercial use only, under CC-BY-NC 4.0. This community demo uses the small version and makes 10 to 30 seconds.

A community member built this demo with the small version of MusicGen. It isn’t Meta’s, it runs on free Hugging Face machines that can be busy, and it costs no credits.

Open on Hugging Face
  • Plays as it’s made
  • 10 to 30 seconds
  • Not for commercial use

Good at moods, not songs

MusicGen writes the instruments and has no idea what a chorus is for, so ask for a sound. The picture is 30 seconds of “Quiet Hours”, solo piano from our generator, about as long as this demo can go.

A spectrogram of 30 seconds of solo piano, with separate bright horizontal lines for each held note.

Name the instruments

The demo’s own example reads “a cheerful country song with acoustic guitars”.

Shorter chunks start sooner

A lower streaming interval, from 0.5 to 2.5 seconds, plays sooner and takes more work.

Same seed, same idea

The seed slider runs from 0 to 10. Move it to get a different result from one sentence.

Half a minute goes quickly

Two more first-30-second pictures from our generator: enough to judge a groove, and short for a video.

An orange waveform of 30 seconds of a lofi beat, with an even pulse of small peaks.

A steady loop

Beats like “Rain Notes” barely change across 30 seconds, which is a fair thing to ask this demo to try.

A spectrogram of 30 seconds of a trap beat, with heavy bright bass along the bottom edge and sharp hits above.

Heavy low end

The bright strip along the bottom of “Cold Bells” is the bass. Listen on proper speakers, since it’s the first thing a small model gets wrong.

This demo, Meta’s demo, and our maker

Meta’s own demo can’t be placed on other sites, so here’s how the three compare. You can open it on Hugging Face.

LengthThis community demo: 10 to 30 secondsMeta’s demo: Up to 120 secondsOur maker: 2 to 4 minutes
SingingThis community demo: NoMeta’s demo: NoOur maker: Yes, or switch it off
Paid use of resultsThis community demo: NoMeta’s demo: NoOur maker: Yes, once you’ve paid
CostThis community demo: FreeMeta’s demo: FreeOur maker: 10 credits for 2 songs on V6

MusicGen questions

Can I use MusicGen music on YouTube or in an ad?

Not for anything commercial. Meta’s code is open under MIT, but the trained weights are CC-BY-NC 4.0, which rules out commercial use. Monetized channels and ads are a poor fit.

Does MusicGen sing?

Not in a way you’d mistake for a person: its makers say it can’t produce realistic vocals. If you need words sung, look at ACE-Step, DiffRhythm or our maker.

Why is this not the official MusicGen demo?

Meta’s demo blocks being placed inside other sites, so this page loads one built by a community member, Sanchit Gandhi. It uses the small MusicGen model and streams its audio.

What if I hear no sound from the demo?

Its author says it works best in Chrome, so try that first. A long wait may also just be the queue on Hugging Face’s shared machines.

Not covered here.Ask us something else

Need music you can publish?

Our maker writes full-length instrumentals. Once you’ve bought any plan or pack, you can use them in paid work.

Make instrumental music

Two free songs, no account needed.