1 week ago
3
Musicians could soon be able to remix the songs that they upload to YouTube thanks to an experimental AI tool currently rolling out to select content creators.
The new tool is built atop YouTube’s Dream Track, which was released last year and enables users to compose songs based on text prompts and by using prerecorded vocals. Charli XCX, Demi Lovato, John Legend, Sia, T-Pain, and Charlie Puth have all signed on for the use of their vocal likenesses.
The new feature acts as sort of a co-producer, allowing the creator to direct the remix and generate a 30-second sample based on the input song and the user’s text prompts. With it, you’ll be able to quickly reimagine a pop tune as a reggaeton bop, or generate a thrash metal version of the 1812 Overture.
“If you’re a creator in the experiment group, you can select an eligible song, describe how you want to restyle it, then generate a unique 30-second soundtrack to use in your Short,” YouTube’ said in its announcement. “These restyled soundtracks will have clear attribution to the original song through the Short itself and the Shorts audio pivot page, and will also clearly indicate that the track was restyled with AI.”
Surely, if you were already going to go through the effort of writing a song, you wouldn’t just compose it in the style that you hear in your head. That’s why the new feature seems made for content creators, as it allows them to soundtrack their Shorts videos with something other than the minor key version of Pink Pony Club.
As with the other Dream Track features, the new remix tool runs on Google’s Lyria large language model, which is trained to generate unique musical scores based on the user’s text prompts (it’s the same idea as image or video generation, just with audio outputs instead).
Google has asserted that all generated tracks, regardless of how much human input there was in their creation, will be clearly and obviously labeled as being made with AI, however, there is no word yet as to whether the company will apply its SynthID watermarking scheme to the new outputs.
English (US) ·