As the music industry undergoes transformative changes with the integration of GenAI tools, Google is amplifying its commitment to AI technology in song and lyric creation. The latest release, MusicFX, represents an evolution of Google’s generative AI text-to-music experiment, MusicLM, introduced last year.
MusicFX offers users the ability to craft tunes up to 70 seconds in length, along with music loops, providing a platform for creativity and exploration. Users can delve into prompts using expressive chips, extending their tracks, and easily download or share their musical creations with friends.
Since its initial launch, MusicFX has witnessed significant traction, with users generating over 10 million tracks. Google’s ongoing efforts to enhance the underlying MusicLM model have resulted in new capabilities, such as higher-quality audio and faster music generation.
Early testing indicates that music samples generated with these updates are now preferred three times more than previous versions.
Google MusicFX and MusicLM
The recent unveiling of MusicFX represents an upgrade to MusicLM, promising users a higher-quality and faster music generation experience. While technically available to select users since December, MusicFX is now generally accessible through Google’s AI Test Kitchen, an app showcasing experimental AI-powered systems from Google’s labs.
MusicFX maintains the user-friendly approach of its predecessor, allowing users to input text prompts describing the desired song. New features include suggestions for alternative descriptor words, providing users with additional creative options. For instance, a prompt like “country style” might offer alternatives such as “rockabilly style” or “bluegrass style.”
Google has implemented certain restrictions to align with copyright regulations. MusicFX filters prompts that mention specific artists or include vocals to avoid potential copyright infringements.
The integration of SynthID, an inaudible watermarking technology developed by Google’s DeepMind division, serves to transparently identify tracks originating from MusicFX. Significant investments have been made in ensuring the safety of training data for Imagen 2, the underlying technology for ImageFX, capable of producing highly realistic images.