Google brings Lyria 3.5 music generation to the Gemini app and API: full songs at 44.1kHz with SynthID watermarking

First, what exactly is Lyria 3.5? It is Google DeepMind's music generation model, first revealed a little over a month ago as a Google Flow Music exclusive — this is its first large-scale arrival inside Gemini. Per Google's developer documentation, Lyria 3.5 accepts text prompts or image inputs and outputs 44.1kHz stereo audio with 'structural coherence': vocals, a timed lyric timeline and full instrumental arrangements. Users get more than a melody loop — they get a complete, multi-minute song with verses, choruses and bridges. Inside the Gemini app, users combine genre selection with prompt input, and can choose whether the track includes vocals or stays instrumental-only; the docs also show Lyria 3.5 supports a new text-lyrics feature that puts songwriting words under user control. For creators who need short loop material, Google keeps the lyria-3-clip-preview model, which always produces 30-second clips for previews, loops and short-video soundtracks.

To appreciate the weight of this release, look at the wider AI music landscape. Over the past two years, Suno and Udio rose to fame on the strength of one-prompt full-song generation, alarming the traditional music industry; after Udio was carved out in 2024 into the DeepMind-backed Uncharted Labs team, Suno became the most-watched independent player, reportedly valued in the tens of billions. Meanwhile copyright litigation has hung over AI music like a sword of Damocles — on July 31 of this year, a Munich court ruled against Suno, finding its use of copyrighted works in training infringing and asserting jurisdiction even over training conducted outside the EU. Google chose this moment for a broad Lyria 3.5 rollout while emphasizing SynthID watermarking — an overt play on the safety-and-compliance card: make AI-generated content technically identifiable and traceable, address regulatory concerns, and extend an olive branch to enterprise customers.

For everyday users, the most tangible change is that making music now only requires the ability to type. In the Gemini app, pick a genre — pop, electronic, hip-hop, folk and more — type a description such as 'a wistful song about a summer beach farewell, female voice, a little jazzy,' and Lyria 3.5 returns a complete track within seconds. For short-video soundtracks or ringtones, the 30-second lyria-3-clip-preview is a better fit. Importantly, generated content carries SynthID — an imperceptible marker embedded in the audio spectrum that human ears cannot hear but detection tools can identify, letting platforms and creators verify whether audio was AI-generated. For platforms like YouTube and TikTok that require AI-content disclosure, this gives AI music an ID card. Google also notes Lyria 3.5 handles multilingual vocals, not just English, opening up a lot of room for global music creation.

Putting Lyria 3.5 inside Gemini also matters strategically for Google. The company has visibly accelerated AI content generation this year — it shipped Gemini 3.7 Flash and 3.8 Flash in August alone — and now it is stuffing music generation into the Gemini app, which counts hundreds of millions of users. That is a classic traffic-hub-plus-model-capability combination. For developers, Lyria 3.5 in the Gemini API means music-teaching apps can embed AI composition demos, video tools can add one-click scoring, and game studios can rapidly generate ambient BGM. Google's pricing has also been aggressive, continuing the Flash family's low-price, high-volume approach to amortize costs through scale and lock developers into its ecosystem. What deserves attention next: how will Suno and Udio respond? Will copyright rulings reshape AI music's business model? And when everyone becomes a composer, will the traditional music industry's rules be rewritten for good?

On a longer timeline, Lyria 3.5's arrival in Gemini may prove to be a milestone. It turns AI music from a geek toy into a national app — hundreds of millions of Gemini users can go from idea to finished track inside a chat without installing anything. AI will not replace songwriters, but it will certainly redistribute the craft of songwriting: arrangement, mixing and vocal synthesis that once required trained musicians now fit in a single prompt. And SynthID's watermark keeps this democratization from sliding toward uncontrolled deepfake music. The real question is the business model: as the marginal cost of generating music approaches zero, value in the music industry migrates from the artifact to live performance, rights and emotional connection — precisely the things AI struggles to replicate. Google's move is pulling that future closer to everyone, right now.