Google DeepMind just launched Lyria 3 in the Gemini app. Type a text prompt (or upload a photo) and you will get a 30-second track with auto-generated lyrics, vocals, and custom cover art. The model is available in eight languages to anyone 18+. YouTube creators worldwide can also access it through Dream Track for Shorts soundtracks. Every track carries a SynthID watermark.
The instrumentals are impressive and can pass for production music: the kind of tracks that underscore voiceovers and get buried under sound effects in corporate videos, podcasts, and social content. The vocals are less convincing and land in the uncanny valley. The 30-second cap is just long enough to make the output useful.
This is still a parlor trick – a very good parlor trick – but a parlor trick nonetheless. A year ago, AI-generated music sounded like a MIDI file having a seizure. Today, Lyria 3 produces tracks with realistic sonority, genre-appropriate arrangements, and coherent song structure. The rate of improvement is mind-boggling. If you extrapolate the quality gains of the past 12 months, the gap between AI-generated music and professional production narrows to a sliver by early next year.
The near-term impact lands squarely on production music, which is a billion dollar market filled with tracks designed to be heard but not listened to. Lyria 3 (and its competitors) can generate this kind of output right now. If you are licensing stock music, Lyria 3 offers an interesting new workflow. The longer-term implications depend on how fast the tracks clear the uncanny valley and how the licensing landscape evolves.
As always, your thoughts and comments are both welcome and encouraged, but please don’t write to tell me that AI will “never” replace human musicians or artists or writers or engineers or composers, etc. I get it. AI is a massive threat and, at the moment, it sucks. On this we can all agree. But remember: we live on the exponential and today AI is the worst it will ever be.
Author’s note: This is not a sponsored post. I am the author of this article and it expresses my own opinions. I am not, nor is my company, receiving compensation for it. This work was created with the assistance of various generative AI models.