AI this, AI thatit seems like every time we cover a new AI breakthrough in the film and video industry another AI innovation is released before we can hit send on the previous article. Good thing all of these pieces are written by AI, right? (Just kidding, I’m a real person… for now at least.)

The concept of AI-powered music generation has always seemed to be one of the many next logical steps in the artificial intelligence take-over of any and all tasks in video production. While there are arguments to be made about how AI might not ever (or at least anytime soon) really replace the more nuanced elements of creative artistic expression—which cinematography, video editing, and music would fall under—it appears that at least the inroads are now finally here.

Meta has released to the public their new AI-powered music generator MusicGen, which promises to allow users the ability to create high-quality, royalty-free music with a simple text description prompt. Let’s take a look at this new AI music app and explore how it might, or might not, eventually take over music for video as we know it.

MusicGen by Meta

Before we dive too deeply into MusicGen by Meta, it’s interesting, or perhaps even a bit commendable, to note that Meta’s music generation model has been released as an open-source project and is being shared throughout both the AI and the music communities. 

Interested users or nervous investors can keep tabs on its developments, contribute to its design, or even eventually make their own variations of it based on the initial neural network. However, for those strictly concerned about what industries this might replace or how it might affect music licensing for video, those questions still remain to be answered for now.

As you can see in the introduction video posted by an AI researcher at Meta, MusicGen is designed to be a simple and controllable music generation model. It can be prompted by both text and melody to create new renditions of your favorite pop classics or create its own melodies with text conditioning and sampling (which you can see in action below).

How It Works

Without leaning too much into the complicated world of the machine learning functions behind the scenes here, what MusicGen is offering is a neural network that can operate off a basic text description or audio input. In a matter of mere seconds, MusicGen can take any prompt you desire and come up with 15 seconds of audio based on your cues.

MusicGen also promises to be quite customizable and exact as you can offer prompts based on beats per minute or provide your own audio clips with the goal of creating different versions or loops. And, if you sign up for the project’s Hugging Face space, you’ll be able to get access to clips up to two minutes.

All exciting and scary stuff, right?

The Future of Music Generation for Video

Still, there are so many questions surrounding this new release. For starters, questions of ethics and legality come to mind. However, according to Meta, MusicGen has been trained only on 10,000 hours of high-quality authorized songs and almost 400,000 instrumental tracks coming from media libraries provided by Shutterstock and Pond5.

As anyone who’s done any amount of video work that requires different types of royalty-free or copyright-free music can attest, the world of music licensing for video is quite complex and tricky to navigate. There’s a lot to be seen about how these AI-powered music loops might actually be usable in the world of video or even just audio production and editing.

But rest assured, by the standards of this fast-moving world of AI these new innovations and uses have been a long time coming. It’s no surprise to see an AI-powered music generator pop up, but it will undoubtedly be surprising to see how it develops and how video pros and the industry overall adapt to these new AI technologies.

How do you see AI-powered music working for film and video? Is MusicGen something you’re excited to try out?

Let us know your thoughts in the comments below.