- TikTok is trialing a major “AI Song” feature that leverages a powerful language model called “Bloom.”
- It’s currently unclear exactly how it works, but it’s believed that some users already have access.
Seven months after introducing Ripple, an AI-powered music creation tool, TikTok has started testing an ‘AI Song’ feature. Ripple can create songs based on a melody the user hums.
This allows users to generate music using text prompts on the platform. Redditors discovered the feature is currently available to a select group of TikTok users in the video-upload section, signaling a targeted approach in this initial phase.
The ‘AI Song’ function harnesses the capabilities of Bloom, a large open-source language model with a staggering 176 billion parameters, adept in 46 natural languages. This puts Bloom on par with OpenAI’s GPT-3.5, a step behind GPT-4, the most powerful language model we currently know.
For now, the feature remains in a testing phase, enabling users to produce music for their TikTok content through simple text inputs. The scale of this testing phase and launch schedule are yet to be disclosed.
The AI Song feature, as explained by TikTok spokesperson Barney Hooper, isn’t exactly an AI song generator in the traditional sense.
Hooper described to The Verge, “It’s not technically an AI song generator — the name is likely to change, and it is currently in testing at the moment. Any music used is from a pre-saved catalog created within the business. In essence, it pairs the lyrics with the pre-saved music, based on three genres: pop, hip-hop, and EDM.”
That’s about as clear as mud, so we need to wait to see how it works.
The integration of AI in music creation on TikTok points to a larger trend in the industry, where the cost-effectiveness and versatility of AI-generated music are becoming increasingly attractive.
However, there are concerns about embedding AI into social media, with a sharp rise in AI ‘deep fake’ images, video, and audio, often used to launch scams, smear public figures, and spread misinformation.
Suzy Loftus, the head of trust and safety at TikTok, highlighted the platform’s commitment to addressing misleading AI-generated content, stating, “We will continue to block manipulated content that could be misleading, including AIGC [AI-generated content] of public figures if it depicts them endorsing a political view.”
This reflects a tricky balancing act between technological innovation and responsible content management – something virtually all social media platforms are grappling with.