YouTube’s teases AI tool that clones famous singers — with their permission

[Collection]

Illustration: Alex Castro / The Verge

Google is testing new generative AI features for YouTube that’ll let people create music tracks using just a text prompt or a simple hummed tune. The first, Dream Track, already seeded to a few creators on the platform, is designed to auto-generate short 30-second music tracks in the style of famous artists. The feature can imitate nine different artists, who’ve chosen to collaborate with YouTube on its development. YouTube is also showing off new tools that can generate music tracks from a hum.

The announcement comes as YouTube attempts to navigate the emerging norms and copyright rules around AI generated music, while also protecting its relationship with major music labels. The issue was brought into sharp relief when an AI-generated “Drake” song went viral earlier this year, and YouTube subsequently announced a deal to work with Universal Music as it establishes rules around AI-generated music on its platform.

YouTube says its Dream Track feature is currently being tested with a “small group of select US creators,” and can produce tracks in the style of nine artists; Alec Benjamin, Charlie Puth, Charli XCX, Demi Lovato, John Legend, Papoose, Sia, T-Pain, and Troye Sivan. In a pair of video demonstrations, we’re shown how the prompt “A ballad about how opposites attract, upbeat acoustic” can create a track in the style of Charlie Puth, or “A sunny morning in Florida, R&B” can be used to make a T-Pain song. The software can generate lyrics, a backing track, and an AI-generated voice in the style of the artist.

The idea is for these tracks to be used with YouTube’s TikTok-style Shorts service. That’s also where YouTube announced a new AI feature called Dream Screen in September, which can generate videos and photos to use as backgrounds.

Away from Dream Track, YouTube has also shown off other Music AI Tools that are designed to let you build out music tracks, often without playing a traditional instrument. In a video demonstration, YouTube shows how the tool can create a saxophone track by combining a hummed melody with the text prompt “saxophone solo.” Other clips show beatboxing being turned into a drum loop, singing becoming an orchestral score, or chords from a MIDI keyboard creating in a vocal choir.

Image: Google

An example user interface for YouTube’s Music AI tools.

YouTube says participants in its Music AI incubator will be able to test these tools later this year.

These new AI tools are powered by a music generation model called Lyria from Google’s DeepMind. In an accompanying blog post from DeepMind, the Google subsidiary says tracks created using Lyria will carry a SynthID watermark that’s inaudible to the naked ear and can be preserved when a track is modified. So even if someone adds more noise to a track, compresses it into an MP3 file, or speeds it up, it should theoretically still be possible to tell that it contains AI-generated audio from Lyria.

The reveal of the new tools comes just days after YouTube announced new content guidelines for AI-generated deepfakes meant to protect people and, ultimately, the platform’s music industry partners.