Lyria 3.5
Lyria 3.5 is Google's newest music generation model, with richer melodies, better lyrics, more expressive vocals, and easier control over tempo and length. This page tracks what changed, where you can actually use it, and how it compares with Lyria 3 — and lets you generate a track here with Lyria 3 Pro.
Generate music on this site
Lyria_3_5_Launch_Showcase.mp4
Wbudowany podgląd demo
Made with Lyria 3.5
Hear what Lyria 3.5 actually sounds like before deciding whether to switch to Lyria 3.5. Every clip below was shared publicly by Google Flow Music, Google Labs, or the artist credited in its title.
"Hasta Que Salga El Sol" — made in Google Flow Music
Press play to listen
"Syncopated Heart" by Eyisha Zyer — made with Lyria 3.5
Press play to listen
Now playing — shared by Google Flow Music
Press play to listen
Retro collage pop — shared by Google Labs
Press play to listen
What's New in Lyria 3.5
Lyria 3.5 is Google DeepMind's most capable music generation model to date, improving melody, lyrics, vocals, and creative control over earlier releases.
Richer Melodic Structures
Lyria 3.5 writes richer, more complex melodic structures that sound more natural across a full track, rather than looping a single idea until the clip ends.
Higher Quality Lyrics
Lyria 3.5 improves prompt adherence and structural awareness in lyrics, so verses and choruses follow the brief you actually wrote instead of drifting off topic.
More Expressive Vocals
Vocals are more realistic and emotionally nuanced, with improved pronunciation — the single most contested area in earlier releases.
Easier Tempo and Length Control
Lyria 3.5 adds explicit BPM control and exportable stems for full-length songs, alongside easier control over tempo and duration, so a 60-second cue or a full three-minute track comes out at the speed you asked for.
Lyria 3.5 Capabilities for AI Music Generation
Lyria 3.5 combines full-length song structure, multi-language vocals, granular prompt control, and SynthID watermarking on every track.
Tracks Up to Three Minutes
Lyria 3.5 reaches about three minutes, long enough for a complete arrangement with intro, verse, chorus, and bridge instead of a short repeating loop.
Vocals in Multiple Languages
Lyria 3.5 generates sung vocals in different languages and specifically targets clearer pronunciation — a known weak point for non-English lyrics.
Prompt Down to BPM and Key
Lyria 3.5 pairs direct BPM control with prompts that accept genre blends, instrumentation, dynamics, key, and a vocal profile covering gender, range, texture, and language.
Song Structure Tags
The Lyria 3 API accepts structure tags such as Verse, Chorus, Bridge, Intro, and Outro, or timestamps, to control how the song progresses.
Exportable Stems for Full Songs
Lyria 3.5 exports separate stems for full-length songs, so a generated track can be taken into a DAW and mixed, layered, or re-arranged instead of staying a flat master.
SynthID Watermark on Every Track
Google applies an inaudible SynthID watermark to every generated track, so it can be identified as AI generated without affecting how it sounds.
Lyria 3.5 vs Lyria 3, Lyria 3 Pro, and RealTime
Lyria 3.5 is the newest model in the family, but it is not simply a renamed Lyria 3 Pro. Here is how the Lyria models differ on length, vocals, and creative control.
| Model | Track length | Vocals | Creative control |
|---|---|---|---|
| Lyria 3.5 (July 29, 2026) | Up to about 3 minutes | Most expressive to date, with improved pronunciation | BPM control plus exportable stems for full-length songs |
| Lyria 3 Pro | A couple of minutes, promptable | Sung vocals in multiple languages | Structure tags and timestamps via the API |
| Lyria 3 Clip | Fixed 30 seconds | Sung vocals in multiple languages | Prompt-level control, no length steering |
| Lyria 2 | Shorter output, instrumental focus | Superseded by Lyria 3 | Superseded by Lyria 3 |
| Lyria RealTime | Continuous, never-ending stream | Instrumental stream | Steered live while the music plays |
How to Use Lyria 3.5
Three steps from a music idea to a downloadable track, whether you are testing it in Flow Music or generating here with the Pro preview.
Write a Detailed Prompt
Name the genre, mood, tempo in BPM, key, instruments, and vocal profile. Specific prompts beat long conflicting ones, and Lyria 3.5 targets exactly this kind of prompt adherence.
Set Length and Vocals
Lyria 3.5 takes length from the prompt itself: ask for a 60-second cue, a half-length edit, or a full three-minute song, and choose instrumental or vocal mode with the lyric language you want.
Generate, Review, Download
Generate the track, listen back, refine the prompt if the structure drifts, then download the audio for your video, podcast, or DAW session. Every Lyria 3.5 export stays watermarked with SynthID.
Should You Switch from Lyria 3 to Lyria 3.5?
Google describes four improvements in Lyria 3.5. Here is what actually changed for vocals, lyrics, and stems before you move a workflow.
Improved: Vocals and Pronunciation
Vocal realism was the loudest complaint about earlier releases, and it is one of the four areas Google explicitly says Lyria 3.5 improves, along with pronunciation.
Improved: Prompt Adherence
Lyria 3.5 lyrics land with better prompt adherence and structural awareness, addressing the recurring problem of dropped lines and repeated phrases in generated songs.
New: Stems You Can Edit in a DAW
Stems now export for full-length songs, so mixing and re-arranging moves into your DAW. Regenerating a section inside the tool is still not supported — generation remains single-turn.
Cennik
Subskrybuj albo płać za użycie dzięki elastycznym planom. Subskrypcję możesz anulować w dowolnym momencie
Starter
- 12 000 kredytów rocznie
- Media do muzyki (Tekst / Teksty / Obraz / Wideo / Audio)
- Pełna generacja wokalu
- Generowanie tekstów
- Generowanie okładki
- Licencja do użytku komercyjnego
- Wsparcie e-mail
- Anuluj w dowolnym momencie
Standard
- 36 000 kredytów rocznie
- Media do muzyki (Tekst / Teksty / Obraz / Wideo / Audio)
- Pełna generacja wokalu
- Generowanie tekstów
- Generowanie okładki
- Licencja do użytku komercyjnego
- Wsparcie priorytetowe
- Anuluj w dowolnym momencie
Pro
- 78 000 kredytów rocznie
- Media do muzyki (Tekst / Teksty / Obraz / Wideo / Audio)
- Pełna generacja wokalu
- Generowanie tekstów
- Generowanie okładki
- Licencja do użytku komercyjnego
- Wsparcie 24/7
- Anuluj w dowolnym momencie
Lyria 3.5 FAQ
Answers on the Lyria 3.5 release, what changed from the previous generation, where it runs, API availability, watermarking, and commercial rights.
Lyria 3.5 is Google DeepMind's newest music generation model, described by Google as its most capable music model to date. It turns a written prompt into a full track with sung lyrics, reaching about three minutes in length, and every output carries an inaudible SynthID watermark.
Google announced Lyria 3.5 on July 29, 2026, rolling it out in Google Flow Music. The model card carries the same publication date, making it roughly five months newer than Lyria 3, which was published in February 2026.
Google lists four improvements: richer and more natural melodic structures, higher quality lyrics with better prompt adherence and structural awareness, more realistic and emotionally nuanced vocals with improved pronunciation, and stronger creative control that now includes BPM control and exportable stems for full-length songs. Google has not published a numeric comparison against Lyria 3, and the model card benchmarks Lyria 3.5 against Lyria 2 instead.
Google Flow Music is the only product listed in the Lyria 3.5 model card, available on the web at flowmusic.google and as an official Google LLC iOS app. The Gemini app runs Lyria 3 rather than Lyria 3.5, free for eligible users with the Pro model on paid Google AI tiers. Developers reach Lyria through lyria-3-clip-preview and lyria-3-pro-preview on the Gemini API and Vertex AI. The generator on this page runs Lyria 3 Pro, the strongest Lyria model currently exposed through the API.
Google Flow Music is the Google Labs product that Lyria 3.5 powers, and the only place Google lists the model today. Formerly Producer AI, it wraps songs, playlists, and music video tools around the model, and runs both on the web at flowmusic.google and as an official iOS app.
Google Flow Music runs Lyria 3.5; the Gemini app still runs Lyria 3. Pick Flow Music if you specifically want the newest model with BPM control and exportable stems. Pick the Gemini app if you want a free 30-second clip, or the Pro model on a paid Google AI tier for longer tracks.
No. As of July 31, 2026 the Gemini API music documentation lists only lyria-3-clip-preview and lyria-3-pro-preview, with no Lyria 3.5 model id, pricing, or rate limits published. Developers building today still integrate Lyria 3.
Google states that it does not claim ownership of the original content generated with Lyria 3.5 in Flow Music, which is not the same as a blanket commercial licence — your rights also depend on the platform terms and plan you generate under. Separately, every Lyria 3.5 track carries a SynthID watermark that is inaudible to listeners but lets the audio be identified as AI generated.
Lyria 3.5 applies Google's generative AI usage policies and filters prompts at generation time, so lyrics naming specific recording artists or reproducing copyrighted words are commonly rejected or rewritten. Rewriting the prompt in your own words is the reliable fix; there is no supported way to bypass the filter.
Switch if vocal quality, pronunciation, and lyric accuracy are your bottleneck, since those are the areas Google says Lyria 3.5 improves. Stay on Lyria 3 Pro if you need API access, since Lyria 3.5 has no published model id, or if your workflow depends on editing generated tracks, which single-turn generation does not support.
Generate a Track with Lyria 3 Pro
Write a prompt, pick vocals or instrumental, and download your track — powered by Lyria 3 Pro, the newest model available through the API while Lyria 3.5 stays inside Google Flow Music.