Lyria 3.5
#2 in Music
« Describe a style, tempo, and lyrics: this AI from Google DeepMind quickly generates a complete song with vocals and instruments. Songs can be up to 3 minutes long and can be exported »
Verified Tool
Similar AI Tools
Suno AI V5.5 
Stable Audio 3.0 
Eleven Music 
Lyria 3.5: full songs with vocals, built on licensed music data
Lyria 3.5 is Google DeepMind's music generation model, able to produce complete songs with verses, choruses, vocals and lyrics from a written prompt. It comes at no extra cost inside Flow Music, Google's music platform, and runs $0.08 per song through the Gemini API. Tracks span 30 seconds to 3 minutes in 44.1 kHz stereo. One detail worth knowing, the prompt language sets the singing language, French and many others included.
Pros
- Structured songs with verses, choruses and bridges
- Sings in the language of your prompt
- Trained on licensed music data
- Tempo, duration and lyrics steered from the prompt
- Composes from images too, up to ten
Cons
- Capped at three minutes per track
- No seed parameter to reproduce a result
- Single-pass generation through the API
Tagged lyrics, numeric BPM and images, what Lyria 3.5 actually reads
Lyria 3.5 generates a complete track from a written description, powered by latent diffusion applied to audio. Paste your own stanzas under Verse and Chorus tags, or let the model write the lyrics and hand them back as text.
In practice, the more numbers in the prompt, the better the song obeys. You ask for 95 BPM, a key of B minor and a breathy female voice, and the arrangement follows your plan section by section. The model also reads up to ten attached images and scores what it sees.
| In your prompt | What the model does with it |
|---|---|
| Named genre and instruments | A faithful arrangement instead of generic sounds |
| Numeric BPM and key | Tempo and harmony locked to your values |
| Verse, Chorus, Bridge tags | Structure honored section by section |
| Up to ten images | Mood and palette translated into music |
| Lyrics in French or another language | Singing in that language, pronunciation included |
Lyria 3.5 and Suno v5.5 tested on the same prompt
Licensed training music and SynthID watermarks, Google's safety card
Every Lyria 3.5 output carries an inaudible SynthID watermark, and Google says its music models are trained on licensed data. Context matters here, since its main rivals among AI music generators, Suno and Udio, face lawsuits from major labels over their training data.
Suno v5.5 keeps the edge on length, up to 8 minutes, and adds voice cloning. Lyria 3.5 plays a different card, traceability and fine structural control (your client's legal team will sleep better).
Lyria 3.5 free in Flow Music, metered through the API
Released in late July 2026 inside Flow Music, the model reached the Gemini app, Google AI Studio and the Gemini API in early September. For developers, a full song costs $0.08, while the older Lyria 3 Clip, now flagged as legacy, charges $0.04 per 30-second clip, handy for testing an idea before the final render. These figures move with Google's release cycle, so the API pricing page remains the one source to trust.
- Flow Music, free, with section-level editing
- Gemini app for chat-based songwriting
- Google AI Studio for first experiments
- Gemini API and Vertex AI for developers
Frequently asked questions
Is Lyria 3.5 free?
Yes, inside Flow Music, where the model is included at no extra cost for every account. Developers go through the Gemini API, billed per generated song, and Google AI Studio works as a sandbox. No dedicated subscription exists for the model itself.
Can you edit a Lyria 3.5 track after generating it?
Not through the API, where generation happens in a single pass with no seed parameter. Inside Flow Music, however, you can replace or extend a given section and request instrumentation changes in plain language. Otherwise, tweak the prompt and generate again.
Does Lyria 3.5 sing in languages other than English?
Prompt language decides the singing language. Lyrics supplied in French come back sung in French, with pronunciation clearly improved over Lyria 3, and the model handles many other languages the same way, with lyrics you provide or ones it writes itself.
Can you use Lyria 3.5 music commercially?
That depends on the current terms of the surface you use, Flow Music, the Gemini app or the API. The SynthID watermark proves a track's AI origin but does not settle rights on its own. Check the Google terms tied to your account before monetizing a song.
Verdict: Three minutes of coherent song for eight cents, with training data on the right side of the courtroom. Video makers, podcasters and music app developers gain a dependable composer, and lyricists get to hear their words sung before ever booking a studio.
★ Featured AI Tools ★
AI Alternatives for
Lyria 3.5
Paid
Freemium
Free
Free
Freemium
Free
Free
