Meta's MusicGen AI can create music based on text prompts

(Image: Reuters)

The Audiocraft team at Meta have released a demo of their MusicGen AI, which can generate short clips of audio based on your text description. You can also use samples of existing melodies as a base.

Also Read | Meta's internal survey shows only 26% employees confident in Mark Zuckerberg's leadership: Report

The way it works is really simple; just head over to the demo page and describe what you want to hear in the audio. In our case, we prompted the system with a simple instruction, "80's synthwave with catchy hooks" and then pressed generate. It took a while, but it eventually gave us a 12 second clip.

[video width="1000" height="200" mp4="https://images.moneycontrol.com/static-mcnews/2023/06/tmpsgn_0mg0.mp4" loop="true"][/video]

We then added another instruction to our prompt by telling the system to generate, "80's synthwave with catchy hooks and jazz melodies".

[video width="1000" height="200" mp4="https://images.moneycontrol.com/static-mcnews/2023/06/tmpomxma0h1.mp4" loop="true"][/video]

Depending on your instructions, it may take a while for the system to generate the clips. In our case, it took nearly two minutes to generate the first sample and nearly three to generate the second.

As you can probably tell, MusicGen AI isn't perfect but it's definitely worth checking out. Ironing out the prompts may lead to even better results.

Also Read | Instagram's Twitter competitor leaked in screenshots

The team used 20,000 hours of licensed music to train the transformer model, which works by predicting the next passage in a piece of music.

Earlier last month, Google opened up its own text-to-music AI system to the public called MusicLM, which works similarly to MusicGen.