• Udio is a new AI platform that generates music from text prompts.
• The platform has drawn interest from angel investors including will.i.am, Common, and Mike Krieger, co-founder of Instagram.
• Unfortunately the creators don’t provide much aside from vague statements when asked what Udio was trained on.

Right now there is an artificial intelligence platform for almost anything you want. Do you want to generate an image of a bird fighting a dragon? There are countless options. What about a long-form essay? That can be done as well.

Now there is another entry into the space – music generation. That comes in the form of Udio, a platform that lets you generate a song using text prompts. The solution has caught the attention of several big-wigs in the industry.

“Backed by a16z, with participation from angel investors like will.i.am, Common, Kevin Wall, Tay Keith, Steve Stoute’s UnitedMasters, Mike Krieger (Cofounder & CTO of Instagram) and Oriol Vinyals (head of Gemini at Google), Udio enables everyone from classically trained musicians, to those with pop star ambitions, to hip hop fans, to people who just want to have fun with their friends to create awe-inspiring songs in mere moments,” reads a press release.

That release claims that users can key in text prompts directing Udio to create a song and in under 40 seconds the user will have a fully mastered track. These tracks can be remixed to create iterations or hone the song.

“There is nothing available that comes close to the ease of use, voice quality and musicality of what we’ve achieved with Udio – it’s a real testament to the folks we have involved,” co-founder and chief executive officer at Udio, David Ding said in a statement.

The company shared samples of Udio’s creations on X and it’s almost impressive.

We say almost because there is one important aspect of Udio that its creators are being cagey about – how it was trained.

Chief operations officer Andrew Sanchez told Musically that Udio is trained on “a large amount of publicly-available and high-quality music” which is, vague.

The COO says that the platform has filters in place to prevent copyright infringement when users create a song, but that says nothing of the songs that Udio used to train its platform.

“And by the way, we totally understand how folks are worried about this. We obviously have tremendous respect for artists and all that sort of thing. But that’s our approach: maybe there’ll be some concern, but we’re prepared to show that there’s a great path forward,” Sanchez told Musically.

The exec doesn’t actually give a straight answer when asked about how Udio was trained and that should be a red flag to everybody intending to use the platform.

Firms creating AI platforms often use datasets they haven’t paid to access in order to train their models. These firms justify this by saying that they need to use copyrighted content otherwise their model would be rubbish.

Earlier this year, OpenAI argued that it needed to infringe on copyright in order to create a viable product. As we said back then, but if you business requires stealing other people’s work to remain relevant, then you may be running something more akin to a criminal operation than a flourishing business.

Udio may avoid the ire of the litigious music industry for now purely because the platform is free to anybody during a beta test. Rest assured though, that once Udio starts charging money, the music studios will come knocking.

Users can create a free account and generate up to 1 200 songs per month. At first Udio will generate a ~30 second long sample of a song, you can then extend it by adding intros, additional sections and an outro. Each time you extend a track it generates a new track with the additional sections, from here you can extend the track further. Udio generates tracks two at a time so you have a choice in which direction you could go.

The creators of Udio see it as a tool to help artists create new tracks and while we can see it going that way, we also see this being used irresponsibly. To its credit, Udio will replace the name of an artist it isn’t working with in prompts and replace it with terms that resemble that artist. When we tried to use Johnny Cash the term was replaced with a slew of other tracks. We were also served a moderation error for one track referencing Cash in a prompt.

We tested Udio out and it created a pretty generic song which is to be expected from an AI, but we just hope that we don’t have a surge in these sorts of tracks plaguing music streaming services soon.

That’s provided the platform isn’t sued into oblivion by music publishers.

[Image – Cristhian Adame from Pixabay]