Meta Launches Generative Music AI called MusicGen

In a bid to have their print on the generative music industry, the parent of world’s largest social network company “Meta” has launched a generative music AI called MusicGen.

Similar to Google’s MusicLM, MusicGen has been trained on two thousand hours of licensed music which include ten thousand high-quality music tracks.

These music tracks have been taken from Shutterstock and Pond5.

It is an open-sourced language learning model that has been developed by Meta’s Audiocraft team.

While you can ask MusicGen to generate music through texts, it also takes melody as input.

With different inputs you can play with it on Hugging face Facebook’s handle at this location.

MusicGen Text Prompts Example

Here is a list of prompt examples that you can try on MusicGen.

1. A romantic piano piece with soft strings and a melancholic melody.
2. An energetic hip-hop track with catchy beats, booming bass, and rap vocals.
3. A mystical and atmospheric instrumental track with ethereal vocals and ambient soundscapes.
4. A lively Latin dance song with salsa rhythms, brass section, and vibrant percussion.
5. A haunting orchestral composition with a haunting choir, dramatic strings, and epic percussion.
6. A groovy funk tune with funky guitar riffs, tight bass lines, and a horn section.
7. A dreamy and atmospheric electronic track with pulsating synths, delicate arpeggios, and soothing vocals.
8. A high-energy pop-rock anthem with powerful vocals, driving guitars, and dynamic drums.
9. A serene and meditative meditation track with gentle piano, soothing nature sounds, and calming synth pads.
10. A suspenseful and dramatic film score with tense strings, ominous brass, and thunderous percussion.

In addition to the above text prompt (or similiar to these ones) if you have any melody that you want to feed, record its audio and upload it.

Uploading melody is an optional section which means you can generate music even with text prompts only.

Once you feed the text prompt and the optional melody audio file, click on the generate button and wait for around 2 min 40 sec to let MusicGen generate a unique piece of music.

You can use MusicGen Locally

Now, In addition to using MusicGen on the hugging face website, you can also use it locally.

All you need is at least 16GB of Ram and a GPU.

The AI is available in four models having the largest model of capacity to generate more complex music.

MusicGen is another milestone in the generative music industry with having a shed by one of the largest companies in the world. I think it will open the door for new music art in the music industry.