Google’s Gemini Launches New AI Music Model, Lyria 3

Google has launched a new AI music model, Lyria 3, in beta as part of its Gemini app, it was announced on Wednesday (Feb 18). Now, users can type in text prompts and upload photos and videos to generate up to 30-second songs in a wide variety of popular styles. All songs are accompanied by AI-generated cover art, courtesy of Google’s image model Nano Banana.

In a press release about the model, Lyria’s stated goal “isn’t to create a musical masterpiece, but rather to give you a fun, unique way to express yourself.” This is the third iteration of the Lyria model, and this one includes three key improvements. Previously, users had to provide their own lyrics and had less creative control over elements like style, vocals and tempo; the tracks were also less musically complex.

The press release adds that Lyria 3 was “designed for original expression, not mimicking existing artists” and has “filters in place to check outputs against existing content,” but it notes that its approach “might not be foolproof.” In that case, users can report tracks that may violate one’s rights.

When asked about the nature of Lyria’s training data, a representative told Billboard that the model is mindful of copyright and partner agreements, and that it only trains on music that YouTube and Google have “a right to use under our terms of service, partner agreements, and applicable law.”

Lyria 3 is clearly blocked from creating songs based on specific artists, songs, books and other protected materials, although through Billboard’s testing of the model, it’s apparent that Lyria still understands the characteristics of these prohibited keywords and can employ those characteristics to generate songs.

After prompting a “love song as if you are Heathcliff from Wuthering Heights singing to Cathy in the style of Kate Bush,” Lyria came back with: “While I can’t create a song that directly uses characters from a specific book or exactly imitates a specific artist’s style, I’ve generated a track that captures the fierce, sweeping drama of your prompt!” It describes the track it made as “dramatic ethereal art-pop…steeped in gothic romanticism…featur[ing]… atmospheric synthesizers.” The result was a song that mentioned Heathcliff and “the lonely moors,” indicating a level of understanding of Bush’s musical style and the location and genre of the famed Emily Brontë novel.

While users are permitted to download the song as an MP3 file or as a video, which includes the music and cover art together, all tracks generated in the Gemini app are embedded with SynthID, an invisible audio watermark that identifies Google AI-generated content.