Photo Credit: Google
Google announces its music generation model Lyria 3 Pro just a month after the release of Lyria 3, letting users make tracks up to three minutes long.
Just a few weeks after the release of Google’s music generation model Lyria 3, the company announced the upcoming Lyria 3 Pro, a model that will enable users to create tracks up to three minutes long. It’s a major upgrade to the Lyria 3 model, which only allows creating tracks up to 30 seconds long.
Besides the longer track length, Lyria 3 Pro will also offer better creative control and customization, according to Google. The model understands track structure far better than its predecessor, so users can specify various elements of the piece they want to create from their prompt—including intros, verses, bridges, and choruses.
“Having explored the vast landscape of music tools, the progress on Lyria 3 is incredible, especially the fidelity and musicality,” said François K, a DJ and producer who collaborated with Google to help ensure Lyria’s usefulness to their workflow. “The fashion in which I use generative AI tools never boils down to ‘one-button-click’ prompting. Instead, it’s becoming a versatile part of my arsenal, allowing me to refine ideas with realism and precision.”
Lyria 3’s music generation capability is available in Google’s Gemini AI app, and the Pro model will roll out to Gemini too, but will only be available to paid subscribers. Lyria 3 Pro will also be included in the Google Vids video editing app and ProducerAI, the latter a genAI music production tool that the company acquired last month. Music generation will also be added to Google’s enterprise tools and the Gemini API.
The company said that it used data from its partners, as well as permitted data from YouTube and Google to train the model, and stressed that the model doesn’t mimic an artist. That said, Google said that if users specify an artist in prompts, it will take “broad inspiration” from that artist in its output.
All tracks created with Lyria 3 and Lyria 3 Pro are marked with SynthID to make clear that AI was used in the track’s creation. That’s important, especially as more DSPs are rolling out tools to identify AI tracks and prevent so-called “slop” from proliferating on their platforms.