• by Colin Kirkland , January 18, 2024

TikTok is testing a generative artificial intelligence (GAI) tool called “AI Song” that allows users to create original music based on text prompts, signifying the social-media app's ongoing investment in music and experimental automation.

The feature provides users with the chance to enter a basic text command that then generates a full song with lyrics, vocals and instrumentals in a matter of seconds. Users can then change the genre of the song created in the app, with options for EDM, Pop, Hip Hop etc.

The company says that the lyrics being generated via AI are powered by a large language model (LLM) called Bloom, which claims to be “the world’s largest open multilingual language model,” operating off a data set that contains over 176 billion parameters.

With Bloom’s ability to generate lyrics in 46 natural languages, TikTok’s lyrical output has the chance to cover a wide range of territory for its global user base of almost 2 billion.

The AI Song feature likely stems from TikTok owner ByteDance's Ripple app, which invites users to play around with a virtual recording studio to create a full track from their own sung or hummed melodies using machine learning. Ripple launched in select markets last year.

With music being the backbone to TikTok's short-form video empire, these AI-powered music creation offerings could influence the way creators develop their own soundtracks to potentially viral videos.

Original AI-generated soundtracks may also open up more creative opportunities to influencers taking part in branded in-app campaigns. Not to mention that brands and creators would be off the hook for copyright infringement.

As SocialMediaToday's Andrew Hutchinson notes: “generative AI has the potential to completely upend the stock content industry, by providing unique, copyright-free content that you can freely use, however you please, and many websites are already using AI-generated images.”

“With tools like this incoming, it seems that stock audio will be another aspect set to be disrupted,” Hutchinson adds.

However, this all depends on how reliable and effective the song quality ends up being.<