YouTube’s Dream Track could be the tipping point for AI-generated music
With a new generative music model and new creator tools, YouTube is looking to take an early lead in AI music. What does that mean for musicians?
YouTube Dream Track. Image: Alphabet
For some time now, AI music generation has been the industry’s Waiting for Godot; fast approaching but never really seeming to arrive. With the release of Lyria and YouTube Dream Track, the suspense might finally be over.
Created jointly by Google DeepMind and YouTube, the two companies – which are both owned by parent company Alphabet – call it their “most advanced AI music generation model to date.” Lyria boasts the ability to create vocals and instrumental textures; write lyrics; transform the timbre and tone of one sound into another; and offer nuanced controls over performance and style.
Sure, the audio quality is grainy enough for a breakfast bowl, but Lyria brings several futuristic generative processes under one roof and slaps a user-friendly interface on top. There’s no denying it’s an important step toward consumer-ready AI music creation.
Those creative tools might have some eager producers intrigued, but it’s the announcement of Dream Track that may ultimately prove to be the big story here.
Dream Track, which is currently only available to a closed group of creators, takes the generative power of Lyria and integrates it into YouTube Shorts – the company’s answer to TikTok. Previously, creators looking to add some music to their videos could choose from a vast library of licensed music but Dream Track is capable of generating an entirely new song using just a few written prompts. You can even pick your own singer, with vocal models from artists including Charlie XCX, John Legend, and T-Pain on offer.
It’s a match made in heaven: 30-60 second clips are not only perfect for the 21st-century attention span; they’re perfect for AI music. Generating short musical clips is what these generative models currently excel at, while creating longer stretches of music still represents a significant technical challenge. Moreover, Lyria’s undeniably lo-fi sound may not hold up on an album release, but YouTube’s betting that consumers might be willing to accept it on a short viral clip played over phone speakers.
With the weight of the world’s largest tech companies behind it, YouTube Dream Track could well be the first use-case for AI-generated music that reaches critical mass. If it does, then the implications for artists are significant, and not necessarily positive.
Getting your song attached to a viral video has become a major part of a modern music career. YouTube’s head of music, Lyor Cohen, recently emphasised its importance to the music industry in a company blog post where he wrote: “Shorts are NEARLY DOUBLING an artist’s total reach, so artists can spend more time doing what they do best: making great music.
“Shorts are the appetizer to the entrée,” he continued. “They are the entry point, leading fans to discover the depth of an artist’s catalogue.”
YouTube Shorts aren’t just a way to get discovered, they’re a way to get paid. As of 1 February 2023, the company introduced revenue sharing on these videos, meaning that if an artist’s music is used in a Short, then they can expect to earn income from any viral success. It’s hard to know exactly how much in royalty payments are generated directly from YouTube Shorts, but the company states that if music is used then half of any revenue generated by a video is allocated to music licensing costs. Some estimates put the income from 1m daily views at $1,157.74 per month, theoretically netting music rights holders $578.87, though few credible sources can confirm this at the time of writing.
Keep in mind that YouTube is the fifth biggest music streaming provider by subscriber market share. Once you factor in the amount of music that is streamed on YouTube by non-subscribers, then the platform is easily one of the biggest on the planet. According to reporting from Rolling Stone and research firm, MIDiA, YouTube has the potential to eventually overtake Spotify as the single biggest funder of the music industry. Last year they paid out $6bn in royalties.
And, just like that, the elephant lumbers into the room. Because there is more to Dream Track than simply finding a good use case for AI music. If implemented at scale, generating music could very well become a significant money saver for YouTube, cutting down the amount of royalties they have to pay out to living, breathing artists.
The copyright of a song’s lyrics and music are one of the primary means through which songwriters earn a living. With Dream Track, these two revenue generators are functionally eliminated.
That isn’t to say the music industry is getting the short end of the stick here. Lyria was almost certainly trained on music owned by the world’s largest record label: Universal Music Group. In August of this year, YouTube announced it was working with UMG to leverage the label’s “roster of talent” for an AI Music Incubator program. It would seem that Lyria is the fruit of that joint effort.
So, when Dream Track pulls from UMG’s vast library of music to generate a new song, we can be sure that the label is getting some sort of compensation and that some slice of that money will eventually trickle down to the artists whose music was used to train the model. Maybe.
However, the only individual musicians directly benefiting from this generative process will be the artists who’ve had their voice modelled. They, depending on the deal struck by UMG, should stand to get some type of royalty payment when their voice is used to generate a song.
Can artists get discovered, build a fan base, and go viral primarily through the ubiquity of their vocal model? Perhaps, but right now, and for a long time to come, the vocal models available on YouTube Dream Track will be drawn from a small, exclusive club of label-backed artists – certainly not from young, emerging singers.
All of this might seem a little alarming. But it’s early days – Dream Track is not yet publicly available, and perhaps users will ultimately prefer to use a song they know rather than generating a hazy one-off.
However, Google DeepMind and YouTube are almost uniquely positioned to move the needle here: both companies are owned by a parent mega-corporation, Alphabet, and this allows the left hand to ‘strategically partner’ with the right hand to maximum benefit. Google DeepMind brings a formidable, and very well-established, AI research program to the table, while YouTube has deep connections to the music, TV, and creator industries. It’s likely how they were able to quickly negotiate and implement all this with Universal Music Group.
Even taken individually, these companies operate at a scale so gigantic that their experiments can reshape large chunks of the creative economy. Working together, they’ve managed to produce Lyria within a timespan of months, not years. I’d say the time to begin worrying was yesterday.
Get the latest news, reviews and tutorials to your inbox.Subscribe