You type "an upbeat pop song about summer" into Gemini, and thirty seconds later you have… a thirty-second clip. Nice, but not a song. Then you hear Google has a "Pro" version that writes full three-minute tracks with verses, choruses, and a bridge — and suddenly the questions pile up: What exactly is Lyria 3 Pro? Do I need a paid Gemini plan? Is there an API? And how do you prompt it so the song actually has structure instead of three minutes of the same loop?
That confusion is understandable, because Google shipped Lyria 3 across at least six different products in the space of a few weeks, each with different access rules.
This guide pulls everything into one place. It is based on Google's official announcements, the DeepMind model card, and the Vertex AI documentation — plus hands-on experience running the model in production behind the Lyria 3 Pro generator on this site. By the end you'll know what the model does, which access path fits you, how to write prompts that control full song structure (including the timestamp technique most articles skip), and where the model still falls short.
What Is Lyria 3 Pro?
Lyria 3 Pro is Google DeepMind's most advanced music generation model. It creates original songs up to 3 minutes long — with intros, verses, choruses, and bridges — from a text prompt, including AI-generated vocals and lyrics. Google announced it on March 25, 2026, about a month after the base Lyria 3 model launched inside the Gemini app.
The headline difference from earlier AI music tools is structural awareness. Most generators let you set a genre and a mood; Lyria 3 Pro lets you architect a song. You can ask for "a moody piano intro that builds into an anthemic chorus, then strips back to an acoustic bridge," and the model understands those as distinct sections rather than one continuous texture.
A few technical facts from Google's own documentation:
- Architecture: Lyria 3 is a latent diffusion system applied to temporal audio latents, trained on Google TPUs — per the official DeepMind model card.
- Output: high-fidelity stereo audio, with vocals in 8 languages (English, German, Spanish, French, Hindi, Japanese, Korean, Portuguese).
- Inputs: text prompts, your own lyrics, and reference images (up to 10) to set the mood — there is no audio-upload or voice-clone input.
- Watermarking: every track is invisibly marked with SynthID, Google's watermark for identifying AI-generated content, and outputs support C2PA content credentials.
- Training data: Google states the model was trained on licensed and permissible data, and outputs are checked against existing content.
Lyria 3 vs Lyria 3 Pro: What's the Difference?
Google ships two variants, and picking the right one saves both time and money:
| Lyria 3 | Lyria 3 Pro | |
|---|---|---|
| Max track length | 30 seconds | Up to 3 minutes |
| Song structure control | Basic | Intros, verses, choruses, bridges |
| Timed / advanced lyric control | Limited | Yes |
| Vocals + lyrics | Yes | Yes |
| Best for | Sketches, social clips, testing ideas | Complete songs, video soundtracks, releases |
Rule of thumb: prototype on Lyria 3, produce on Lyria 3 Pro. A 30-second clip is the cheapest way to check whether your genre, mood, and vocal direction actually work before you spend a full-song generation on it. If the clip nails the vibe, promote the same prompt to Pro and add structure. (Both variants are covered in more depth on our Lyria 3 page.)
Where Can You Use Lyria 3 Pro? (6 Access Paths Compared)
This is where most coverage gets vague, because availability differs by product. Here is the full map as of July 2026, based on Google's rollout announcements:
| Access path | Who it's for | Access model |
|---|---|---|
| Gemini app | Casual creation on phone/web | Paid Gemini subscribers |
| Google Vids | Workspace users scoring videos | Workspace customers, Google AI Pro & Ultra |
| Google AI Studio / Gemini API | Developers building apps | Pay-per-generation API |
| Vertex AI | Enterprise / scaled workloads | Public preview, API + Media Studio |
| ProducerAI | Music-first creation | Free and paid tiers, global |
| Credit-based studios (like this site) | Creators who want credits, not subscriptions | Pay-as-you-go credits, no Google Cloud setup |
How to choose, in one pass:
- If you already pay for Gemini, just use the app — it's the fastest path.
- If you want to build Lyria into a product, start in Google AI Studio, then graduate to Vertex AI when you need scale and enterprise controls.
- If you want full-length Lyria 3 Pro tracks without a monthly subscription or cloud console setup, a credits-based studio is the lowest-friction route — you can generate a track with the Lyria 3 Pro generator from a single prompt box, no Google account plumbing required.
How to Prompt Lyria 3 Pro (the Official Formula)
Google published a dedicated prompting guide for Lyria 3 Pro, and its core formula is worth memorizing:
[Genre & style] + [Mood] + [Instrumentation] + [Tempo & rhythm] + [Vocal style & language] + [Lyrics]What each slot does:
| Component | What to write | Example |
|---|---|---|
| Genre & style | Primary category, era references help | "classic Bossa Nova meets modern R&B" |
| Mood | Emotional intent | "intimate, warm, deeply affectionate" |
| Instrumentation | Name the key instruments | "nylon-string guitar, warm electric piano" |
| Tempo & rhythm | Descriptive pace, not just BPM | "slow, swaying tempo" |
| Vocal style & language | Gender, texture, delivery, language | "soft, breathy female vocalist in French" |
| Lyrics | A theme, or exact lyrics after Lyrics: | "a love song about a cross-cultural connection" |
Three flags that change everything:
- Instrumental tracks: end your prompt with the word
Instrumental.— no vocals will be generated. - Your own lyrics: write
Lyrics:followed by your exact words, and the model sings them. - Duets: describe multiple singers ("a smooth male vocalist in English and a breathy female vocalist in French") and Lyria arranges them together.
The Timestamp Technique: Full Control Over Song Structure
This is Lyria 3 Pro's most underused feature, and the main thing that separates it from every "genre + mood" generator. You can script the entire arrangement with timestamps:
[00:00] Begin immediately with a massive gospel choir singing an uplifting harmony.
[00:15] A heavy, modern hip-hop drum beat and deep 808 bassline drop in.
[00:30] Male lead vocalist begins rapping a confident verse.
[01:10] Transition into a triumphant chorus with full choir and brass.
[01:50] Strip back to a quiet, emotional bridge on organ.
[02:10] Full beat and choir return for the final chorus.
[03:00] End on a sustained choir chord.That example comes straight from Google's own guide — and it works because Pro was trained to understand section-level composition, not just texture. Use timestamps whenever you're scoring a video with known scene changes, or when you want a guaranteed bridge instead of hoping one appears.
Low-friction check before you commit: run your genre/mood/vocal description as a 30-second Lyria 3 clip first. If the sound is right, paste the same description into a Pro generation and add the timestamp skeleton. You'll waste far fewer full-song generations this way.
Lyria 3 Pro vs Suno: Which Should You Use?
The comparison everyone searches for. There's no universal winner — the two tools optimize for different things:
- Lyria 3 Pro's strengths: structural control (timestamps, section prompting), high-fidelity stereo output, image-guided generation, and deep integration with Google's ecosystem (Gemini, Vids, Vertex AI). It executes a precise brief faithfully — which is exactly what branded content and video scoring need.
- Suno's strengths: a free tier, a longer track record with hobbyist songwriters, stem downloads on paid plans, and a more "surprising" creative character that many users enjoy for songwriting exploration.
- The honest answer: if you want to architect a specific song for a specific purpose, Lyria 3 Pro's structure control is currently unmatched. If you want to discover songs cheaply and remix them, Suno's ecosystem is friendlier.
If you just want to hear the difference for yourself, the fastest route is to run one prompt through the Lyria 3 Pro generator and the same idea through Suno's own app, then compare the two tracks side by side.
How Much Does Lyria 3 Pro Cost?
Google doesn't publish one unified price for Lyria 3 Pro — cost depends entirely on the access path:
- Gemini app: included with paid Gemini subscriptions (it rolled out "starting with paid subscribers," with daily generation limits that vary by tier).
- ProducerAI: available globally on both free and paid tiers.
- Gemini API / Vertex AI: metered, pay-per-generation pricing while the model is in public preview — per-track rates are listed on Google's current rate cards and have been changing during the preview period, so check them before budgeting a project.
- Credit-based studios: you buy a credit pack once and spend a small number of credits per track — no subscription. This is generally the cheapest way to try full-length Pro generations if you only need a handful of songs; see our pricing page for the exact credit costs on this site.
One cost-shaped detail worth knowing: a Lyria 3 Pro generation produces a complete ~3-minute song in a single request. You don't pay per-30-seconds the way you might assume from the base model — which makes Pro surprisingly economical per minute of finished audio.
Limitations You Should Know Before You Commit
An honest list, sourced from Google's documentation and our own production use:
- 3-minute ceiling. There is no way to generate a 5-minute progressive track in one pass. For longer pieces you'll need to generate sections and edit them together.
- No audio input. You can guide generation with text, lyrics, and reference images — but you cannot upload a melody, a hum, or an existing track to continue. No voice cloning either.
- No stem export from the model. Lyria returns a finished stereo mix. If your workflow needs isolated vocals/drums for mixing, you'll need a separate stem-splitting step.
- 8 vocal languages. Broad, but not universal — vocals outside the supported list (English, German, Spanish, French, Hindi, Japanese, Korean, Portuguese) aren't officially supported.
- Content checks can block prompts. Outputs are screened against existing content, and prompts referencing real artists' styles too directly may be refused — a deliberate consequence of Google training on licensed data.
- Preview status. On Vertex AI the model is still labeled public preview, which means specs and rates can change.
- Every track is watermarked. SynthID is always embedded. It's inaudible and doesn't affect quality, but AI-detection tools can identify the track as generated — worth knowing if provenance matters to your use case.
None of these are dealbreakers for the model's core use cases — social content, video soundtracks, demos, and full song drafts — but knowing them up front prevents the most common disappointments.
FAQ
Is Lyria 3 Pro free? Not directly from Google — the Gemini app requires a paid subscription and the API is pay-per-generation. ProducerAI exposes it on a free tier, and credit-based generators like this site let you pay only for the tracks you generate.
Can Lyria 3 Pro generate vocals and lyrics?
Yes. It writes and sings original lyrics from a theme, or performs the exact lyrics you provide via the Lyrics: prefix, in 8 languages, including duets.
How long are Lyria 3 Pro songs? Up to 3 minutes, with real song structure (intro/verse/chorus/bridge). The base Lyria 3 model generates 30-second clips.
Is there a Lyria 3 Pro API? Yes — through the Gemini API, Google AI Studio, and Vertex AI (public preview). Third-party platforms also expose the model through their own APIs and UIs.
Can I use Lyria 3 Pro music commercially? Usage rights are governed by the terms of the platform you generate on (Google's terms of service and generative AI policies for Google surfaces). Review the current terms for your access path before commercial release — policies during the preview period can change.
Does Lyria 3 Pro work from images? Yes — you can attach reference images (up to 10) and the model uses them to set the story, mood, and style of the track.
Bottom Line
Lyria 3 Pro is the first mainstream music model that treats a song as structure rather than texture: three minutes, real sections, controllable vocals in 8 languages, and a prompting language (including timestamps) that gives you director-level control. Its trade-offs — no stems, no audio input, a 3-minute cap, always-on watermarking — are real but manageable.
The fastest way to form your own opinion takes about two minutes: write one sentence describing a song you actually want — genre, mood, one instrument, one vocal direction — and run it through the Lyria 3 Pro generator. If you like the vibe at 30 seconds on the base model, spend the credits on the full 3-minute Pro version.
Sources
Primary sources used for the factual claims in this article. Note: Lyria 3 Pro is partly in public preview — capabilities and rates may change after publication (July 2026).
- Google: Lyria 3 Pro announcement — release date, 3-minute tracks, product availability
- Google: Build with Lyria 3 (developer announcement) — API availability
- Google DeepMind: Lyria model page — capabilities, model family
- Google DeepMind: Lyria 3 model card — architecture, training, SynthID
- Google Cloud: Lyria 3 and Lyria 3 Pro on Vertex AI — Vertex preview, inputs, C2PA
- Google Cloud: Ultimate prompting guide for Lyria 3 Pro — prompt formula, timestamps, languages
- TechCrunch: Google launches Lyria 3 Pro — launch coverage
