YouTube’s generative music experiments come at a tense time for artists and AI

YouTube Music is passing the mic to AI with a new feature that will allow users to generate music in the style of pop stars like Sia, John Legend, and T-Pain—with the permission of the artists.

The tool, which draws on a Google DeepMind music generation model called Lyria, is designed to produce unique backing tracks for creators on the TikTok-esque YouTube shorts platform. It comes at a fraught time in relations between AI engineers and the artists and writers whose work their models rely on.

YouTube’s Dream Track feature can auto-generate 30-second audio clips based on a text prompt like “a ballad about how opposites attract, upbeat acoustic.” Nine artists signed on to have their vocal and instrumental likenesses cloned: Alec Benjamin, Charlie Puth, Charli XCX, Demi Lovato, John Legend, Papoose, Sia, T-Pain, and Troye Sivan.

Another YouTube tool, Music AI, can turn a hummed tune into a guitar riff or convert a song to a different genre. Google will also use its SynthID watermark to apply imperceptible-to-the-ear differentiation in tracks produced by the tools.

Battle of the bands: Creative and social platforms have been experimenting with generative AI music cloning since at least as far back as the 2020 release of OpenAI’s Jukebox program. But like much generative AI output, artificial music has hit new heights of passable realism in the last year or so, as exemplified by a fake Drake song that went viral in April.

That’s naturally led to tension among artists, the record industry, and tech companies looking to monetize this type of AI. Universal Music Group, for instance, sued Anthropic last month over its chatbot’s unauthorized distribution of song lyrics. And just last week, superstar Bad Bunny posted an angry missive about an AI track that used his vocal likeness.

Issues like these led generative startup Stability AI’s head of audio, Ed Newton-Rex, a composer himself, to resign from his post this week, explaining in an op-ed for Music Business Worldwide that he disagrees with “the company’s opinion that training generative AI models on copyrighted works is ‘fair use.’”

Courting the music biz: YouTube Music has been working to position itself as a better alternative for artists to use AI to their benefit by appealing directly to them and record labels. In August, the company collaborated with Universal Music Group on a set of principles for harnessing music-generating AI responsibly.

This week, YouTube also unveiled a set of rules around deepfake videos uploaded to the platform, which notably only allow labels and distributors of artists working with YouTube on AI music experiments to request that the deepfakes be removed.

If YouTube Music is able to pull off a healthy partnership with the music industry, it could prove a valuable tool for the company to compete with the likes of TikTok, whose parent company unveiled its own music generator earlier this year. But norms around copyrights and revenue splitting are still being hotly debated, and tensions around the new technology only seem to be ramping up.