Rage Against the Machine Learning
With a simple prompt, you can generate 30 seconds of something like music.
The American poet Henry Wadsworth Longfellow called music “the universal language of mankind.” Is that still true when the so-called music is being generated by a probabilistic robot instead of a human? We’re about to find out. Google has announced its latest Lyria 3 AI model is being deployed in the Gemini app, vastly expanding access to AI music generation.
Google DeepMind has been tinkering with Lyria for a while now, offering limited access in developer-oriented products like Vertex AI. Lyria 3 is more capable than previous versions, and it’s also quicker to use. Just select the new “Create music” option in the Gemini app or web UI to get started. You can describe what you want and even upload an image to help the robot get the right vibe. And in a few seconds, you get music (or something like it).
In case there was any uncertainty about whether Lyria tracks still counted as a human artistic endeavor, worry not! Unlike past versions of the model, you don’t even have to provide lyrics in your prompt. You can be vague with your request, and the model will create suitable lyrics for the 30-second song. Although with that limit, “jingle” might be more accurate.
In addition to the track, each music creation job will come with an album cover-style image created by the Nano Banana model. Gemini will also have a pre-loaded set of AI tracks that you can choose to remix to your heart’s content. The Lyria 3 tools are also coming to Google’s Dream Track toolkit for YouTube Shorts, which will pair nicely with the Veo AI video options.
So what kind of tracks can you expect Gemini to spit out? Google has provided some examples:
“Sweet Like Plantain“
Prompt: I’m feeling nostalgic. Create a track for my mother about the great times we had as kids and the memories of her home-cooked plantains. Make it a fun afrobeat track with a true African vibe.
“Motown Parody“
Prompt: Quintessential 1970s Motown soul. Lush, orchestral R&B production. Warm bassline with melodic fills, locked into a steady drum groove with crisp snare and tambourine. Vintage organ harmonic bed. Three-piece brass section. Gritty, gospel-tinged male tenor lead.
“Pop Flutter“
Prompt: Wistful and airy. Soft, breathy female vocals with intimacy. Rapid-fire drum and bass rhythm, low-passed and softened. Deep, warm bass swells. Dreamy electric piano chords and subtle chime textures. Rainy city vibes.
“Sea Shanty“
Prompt: An authentic A capella Sea Shanty featuring a robust male choir singing in a traditional call-and-response format. The piece is entirely vocal, relying on synchronized foot-stomps on a wooden deck and sharp handclaps to provide the rhythmic pulse. The lead is a weathered male baritone with a gravelly timbre who sings the narrative ‘chant’ lines. He is immediately answered by a powerful male choir singing in rich, rugged harmony on the ‘response’ lines. The voices are recorded with a natural room reverb that simulates the acoustic environment of a wooden ship’s deck, giving the vocals a resonant, atmospheric quality. The performance is energetic and driving, with the choir leaning into the rhythm of the stomps to create a sense of focused, communal effort. There are no instruments, only the layered textures of collective male voices spanning tenor, baritone, and bass ranges, all contributing to a confident, monolithic sound.
Sour notes
AI-generated music is not a new phenomenon. Several companies offer models that ingest and homogenize human-created music, and the resulting tracks can sound remarkably “real,” if a bit overproduced. Streaming services have already been inundated with phony AI artists, some of which have gathered thousands of listeners who may not even realize they’re grooving to the musical equivalent of a blender set to purée.
Still, you have to seek out tools like that, and Google is bringing similar capabilities to the Gemini app. As one of the most popular AI platforms, we’re probably about to see a lot more AI music on the Internet. Google says tracks generated with Lyria 3 will have an audio version of Google’s SynthID embedded within. That means you’ll always be able to check if a piece of audio was created with Google’s AI by uploading it to Gemini, similar to the way you can check images and videos for SynthID tags.
Google also says it has sought to create a music AI that respects copyright and partner agreements. If you name a specific artist in your prompt, Gemini won’t attempt to copy that artist’s sound. Instead, it’s trained to take that as “broad creative inspiration.” Although it also notes this process is not foolproof, and some of that original expression might imitate an artist too much. In those cases, Google invites users to report such shared content.
Lyria 3 is going live in the Gemini web interface today and should be available in the mobile app within a few days. It works in English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese, but Google plans to add more languages soon. While all users will have some access to music generation, those with AI Pro and AI Ultra subscriptions will have higher usage limits, but the specifics are unclear.
Ryan Whitwam is a senior technology reporter at Ars Technica, covering the ways Google, AI, and mobile technology continue to change the world. Over his 20-year career, he’s written for Android Police, ExtremeTech, Wirecutter, NY Times, and more. He has reviewed more phones than most people will ever own. You can follow him on Bluesky, where you will see photos of his dozens of mechanical keyboards.
