In an age when the boundary between human creativity and machine output grows harder to locate, Google has brought its most advanced music-generation model, Lyria 3.5, into the hands of ordinary users through the Gemini app — allowing anyone to conjure a full song from a sentence or a photograph. The move arrives not in a vacuum but against a backdrop of genuine disruption: AI-generated music already constitutes nearly half of daily uploads on some streaming platforms, and the fraud that has followed raises questions about authenticity, economics, and what it means to make something. Google's
Google Expands Lyria 3.5 AI Music Model to Gemini App for All Users
Every track carries an invisible marker meant to flag it as AI-generated
Mark
So Google just made it possible for anyone to generate a full song. What does that actually change?
Mimi
The barrier drops dramatically. Before, you needed access to Flow, which was specialized. Now it's in Gemini, which millions of people already use. You describe what you want—a lo-fi beat, a folk ballad, whatever—and you get a complete track with structure, not just a loop.
Luke
But can you edit it after it's made?
Mimi
No. That's the constraint. You generate, you get what you get, or you try again.
Mark
Why does the timing matter so much? Why is this announcement happening "now"?
Mimi
Because AI music has flooded streaming platforms. On Deezer alone, 44 percent of daily uploads are AI-generated. A lot of those streams were fake—artificially pumped numbers.
Luke
So Google is releasing this tool into a space where AI music is already a credibility crisis.
Mimi
Exactly. Which is why every track gets a SynthID watermark—an invisible marker saying "this is synthetic."
Mark
Does that actually solve the problem?
nLuke: It identifies the content, but it doesn't prevent fraud. It doesn't stop someone from uploading a watermarked track and running fake streams through it. It's transparency, not enforcement.
Mimi
True. But it's also a signal. Google is saying: we're not hiding what this is.
Mark
And developers can use this too?
Mimi
Through the Gemini API, yes. So music generation moves beyond Google's own apps into whatever third-party services want to build it in.
Der Puls
- Lyria 3.5 marks a genuine technical threshold — where earlier AI music tools looped short fragments, this model builds complete songs with verses, choruses, bridges, and nuanced vocal performances.
- The rollout is immediate and broad: any Gemini user can now generate custom tracks from text or photos, with templates for jingles, video backgrounds, and ringtones, while developers gain access through the Gemini API.
- The stakes are sharpened by what is already happening on streaming platforms — AI-generated tracks make up 44 percent of daily uploads on Deezer, and a significant share of their play counts have been found to be fraudulently inflated.
- The music industry is watching with alarm as automated systems game platform economics, turning AI generation into an instrument of financial manipulation rather than creative expression.
- Google's primary safeguard is the SynthID watermark, an invisible digital signature embedded in every Lyria-generated track — a transparency measure whose real-world enforcement, durability, and cultural weight remain unproven.
In an age when the boundary between human creativity and machine output grows harder to locate, Google has brought its most advanced music-generation model, Lyria 3.5, into the hands of ordinary users through the Gemini app — allowing anyone to conjure a full song from a sentence or a photograph. The move arrives not in a vacuum but against a backdrop of genuine disruption: AI-generated music already constitutes nearly half of daily uploads on some streaming platforms, and the fraud that has followed raises questions about authenticity, economics, and what it means to make something. Google's answer, for now, is a watermark — an invisible signature that names the machine, even if it cannot yet govern what the machine has set in motion.
Google has moved Lyria 3.5, its most capable music-generation model, out of a specialized filmmaking tool and into the Gemini app, where any user can now type a description or upload a photo and receive a complete, structured song in return. The technical distance from earlier versions is meaningful: previous Lyria models produced looping melodic fragments, the audio equivalent of a GIF. Lyria 3.5 builds full arrangements — verses, choruses, bridges, vocal performances or pure instrumentation — and better interprets what a user is actually asking for. Templates inside Gemini guide common use cases, from video background music to personalized ringtones, with options for genre, length, and vocal style. One limitation holds: if a generated track misses the mark, there is no mid-process editing — only starting again.
The release lands in contested territory. AI-generated music has arrived on streaming platforms not as a trickle but as a flood, with such content comprising 44 percent of daily uploads on Deezer alone. The volume might be absorbed if the content were straightforwardly legitimate, but a substantial portion of the streams these tracks accumulated proved fraudulent — play counts artificially inflated by automated systems designed to extract money from platform royalty structures. The music industry has observed this with mounting unease.
Google's stated answer is the SynthID watermark, an invisible digital signature embedded in every track Lyria produces, designed to identify the content as machine-made. It is a gesture toward transparency and accountability in a space that has outpaced the frameworks meant to contain it. Whether streaming platforms will actively enforce detection, whether the watermark can survive deliberate removal, and whether listeners will ultimately care — none of these questions are settled. Google is also releasing Lyria 3.5 through the Gemini API for developer integration, which means the capability will soon propagate well beyond Google's own products, carrying both its creative possibilities and its unresolved tensions into a much wider world.
Google has pushed its most capable music-generation model into the hands of everyday users. Lyria 3.5, previously confined to a specialized filmmaking tool called Flow, is now live inside the Gemini app for anyone with access. The shift is straightforward in its ambition: type a description or upload a photo, and the system will compose an entire song—complete with verses, choruses, bridges, and either human vocals or pure instrumentation.
The technical leap matters. Earlier versions of Lyria would loop a short melodic fragment, the musical equivalent of a GIF. Lyria 3.5 builds outward. It generates fuller arrangements, more nuanced vocal performances, and better understands what you're actually asking for. Inside Gemini, the interface offers templates for common tasks: background music for a video, a custom jingle, a personalized ringtone. Users can specify a genre, choose between vocal and instrumental, and decide whether they want a thirty-second clip or something longer. The system handles the rest.
There is a constraint worth noting. Once a track generates, you cannot edit it mid-process. If the result misses the mark, you start over. Google is also opening Lyria 3.5 to developers through the Gemini API, meaning the music-generation capability will soon appear inside third-party apps and services, not just Google's own ecosystem.
The timing is fraught. AI-generated music has become a flood. On Deezer, the streaming platform, AI-composed tracks now account for 44 percent of all daily uploads. The volume alone would be manageable if the content were legitimate, but a significant portion of the streams these tracks accumulated turned out to be fraudulent—artificially inflated play counts, often generated by automated systems designed to game the platform's economics. The music industry has watched this unfold with alarm.
Google's response is the SynthID watermark, an invisible digital signature embedded in every track Lyria generates. The watermark is meant to flag the content as synthetic, to mark it as something made by machine rather than human hands. It is a transparency mechanism, a way of saying: this is not a person's work. Whether platforms will enforce detection of these markers, whether listeners will care, whether the watermark itself proves robust against removal—these remain open questions. For now, Google is betting that identification, at least, is a step toward accountability in a space that has moved faster than the guardrails meant to govern it.
Bemerkenswerte Zitate
Lyria 3.5 delivers richer arrangements, more expressive vocals, and better prompt comprehension than previous versions— Google (on model capabilities)
Möchten Sie die ganze Geschichte?
Das Original lesen bei Digital Trends ↗
Technology