Lyria 3 Pro: 5 Prompts, Pricing, and What It Can't Do

Two days ago, OpenAI killed Sora — their video generator was burning $15 million a day and Disney walked away from a billion-dollar deal. One day later, Google dropped Lyria 3 Pro, an AI music model that can generate 3-minute songs with vocals, choruses, and bridge sections.

The timing wasn’t subtle. And the message was clear: creative AI isn’t dead. It just changed addresses.

But here’s what matters to you: Lyria 3 Pro is the first AI music tool that actually understands song structure. You can tell it “start with a moody piano intro, build into an upbeat chorus, then wind down with an acoustic bridge” — and it gets it. Previous AI music tools just gave you a blob of sound with no beginning, middle, or end.

Let’s walk through exactly how to use it, what it costs, and where it falls short.

What Lyria 3 Pro Actually Is

Lyria 3 Pro is Google DeepMind’s latest music generation model. It launched on March 25, 2026, and it’s available through the Gemini app (the one you might already use for chatting with Google’s AI).

The big upgrade over the previous Lyria 3: track length went from 30 seconds to 3 full minutes. That’s enough for a real song — intro, verse, chorus, bridge, outro. Not just a loop or a jingle.

It generates 48kHz stereo audio (high quality), handles vocals and instrumentals, and can even write lyrics for you in English, Spanish, French, and Japanese.

One thing that’s genuinely unique: it creates cover art for every track automatically. No other AI music tool does this. Small thing, but it’s a nice touch if you’re making content for YouTube or social media.

Every track gets an invisible SynthID watermark — you can’t hear it, and it survives compression, speed changes, even adding noise. It’s how platforms can tell the music was AI-generated. You can actually upload a track to Gemini and ask “was this made by AI?” to check.

How to Use Lyria 3 Pro (Step by Step)

Getting started takes about two minutes. Here’s the process:

1. Open the Gemini App

Go to gemini.google.com or open the Gemini mobile app. You’ll need a paid subscription — Lyria 3 Pro isn’t available on the free tier. More on pricing below.

Click or tap the Tools icon, then select “Create music.” This switches Gemini into music generation mode.

3. Write Your Prompt

This is where the magic happens. And the more specific you are, the better the results.

A basic prompt works:

“Create a chill lo-fi hip hop track with soft piano and vinyl crackle”

But a detailed prompt works way better:

“Create a 2-minute indie folk song with acoustic guitar, soft female vocals, and light percussion. Start with a fingerpicked guitar intro for 15 seconds, then a verse with lyrics about driving through small towns at night, build into an uplifting chorus, and end with the guitar fading out.”

You can control:

Genre and era (“1980s synthwave,” “modern R&B”)
Instruments (“acoustic guitar, upright bass, brushed drums”)
Tempo and mood (“120 BPM, melancholic but hopeful”)
Vocal style (“raspy male vocals,” “ethereal female harmonies”)
Song structure (“intro, verse, chorus, verse, chorus, bridge, outro”)
Key and scale (“A minor, pentatonic scale”)

Power user tip: Specify structure in bar counts for much more precise results. One user reported that prompting with “BPM120, lo-fi hip hop, intro 8 bars, verse 16 bars, chorus” dramatically increases accuracy compared to vague structure descriptions.

Another trick: Upload a reference image with your prompt. Lyria uses the image as context for both the lyrics and the auto-generated cover art. A sunset beach photo produces different lyrics and vibes than a rainy city street.

4. Add Custom Lyrics (Optional)

If you want specific words, you can include them in your prompt with section tags:

[Verse]
Driving down these empty roads at night
Headlights cutting through the fog
Every town looks the same but feels different

[Chorus]
And I keep moving, keep moving on

You can even add timestamps like [0:00 - 0:15] for precise control over where each section starts.

Pro tip: Write your prompt in a different language and Lyria generates lyrics in that language. A French prompt gives you French lyrics. Works for English, Spanish, French, and Japanese.

5. Listen and Download

Gemini generates the track (this takes 30-60 seconds — it’s not instant). You’ll get a playable preview with that auto-generated cover art. Download as MP3 or WAV.

That’s it. No audio engineering degree required.

What You Can Actually Make With It

Based on what the community has been sharing since launch, here’s what works well:

Background music for videos — YouTube intros, podcast backgrounds, social media reels. This is the strongest use case. One Japanese creator who’s produced 10,000+ videos called it out specifically: “Lyria creates copyright-free music matched to your video’s atmosphere in seconds.”
Children’s content — A Turkish tech creator typed one sentence and got a complete kids’ song in 30 seconds — lyrics, melody, and cover art included. His take: “If you gave this to an agency, how many days and how much money would they ask for?”
Ambient and mood music — Lo-fi study beats, ambient work soundscapes, chill evening vibes. Japanese users specifically noted ambient music as a strong suit: “You could use these commercially right away without issues.”
Cinematic scores — One user posted a detailed Hans Zimmer-style prompt with tempo changes and key progressions, and the result was genuinely impressive orchestral work.
Genre experiments — People have made 1980s Japanese city pop, Yemeni traditional music, opera about garbage bags (combined with Suno and video AI), and video game soundtracks.

And here’s a take that surprised a lot of people: a hobbyist audio engineer with 15+ years of experience said Lyria’s sound design is “much fuller, spatially embedded” than both Suno and Udio. The composition might not always be as creative, but the actual audio quality — the way instruments sit in the mix, the spatial depth — is a clear step up.

Pricing: What Does It Cost?

Here’s the part that frustrates some people — Lyria 3 Pro has no free tier.

Plan	Daily Track Limit	Monthly Cost
Free Gemini	No access	$0
Google AI Plus	10 tracks/day	~$20/mo
Google AI Pro	20 tracks/day	~$20/mo
Google AI Ultra	50 tracks/day	~$250/mo

If you’re already paying for Gemini, you get Lyria 3 Pro included — no extra charge. So it’s really a question of whether you’re already in Google’s ecosystem.

For developers: Lyria 3 Pro is also available through the Gemini API, Google AI Studio, and Vertex AI. The API version currently generates tracks up to about 2 minutes (slightly shorter than the 3-minute app version). SDKs exist for Python, JavaScript, Go, Java, and C#. API pricing works out to roughly $0.08 per song — which is 5x more expensive than Suno’s subscription rate ($0.015/song at the $30/mo tier). But the pay-per-generation model is actually better if you only need a few tracks per month.

There’s also ProducerAI — a music production tool Google acquired (the team behind Riffusion). It works with Lyria 3 Pro and has both free and paid tiers. If you want more of a DAW-like experience with AI assistance, this is the route.

What Lyria 3 Pro Can’t Do

And this matters, because every review seems to gloss over the limitations.

No multi-turn editing. You can’t say “make the chorus louder” or “swap the guitar for a synth.” Every generation is a one-shot deal. If you don’t like it, you prompt again and get something completely different. There’s no seed control, no versioning, no iteration.

No stem export. You get one mixed track. You can’t separate the vocals, drums, bass, and melody into individual files. Suno does this. Udio does this. Lyria doesn’t.

No inpainting. If the verse sounds great but the chorus is weak, you can’t regenerate just the chorus. Udio lets you do this. With Lyria, it’s all or nothing.

No voice cloning or input. You can’t feed it a melody, hum a tune, or upload a reference track. It only takes text.

Lyrics in four languages only. English, Spanish, French, and Japanese. If you need Korean, Portuguese, German, or anything else — you’re out of luck for now. And even in supported languages, pronunciation isn’t perfect — Japanese users reported that vocal pronunciation “felt a bit off” in their tests.

It can’t mimic specific artists. Safety filters block requests like “sing like Adele” or “in the style of The Weeknd.” Artist names are treated as “broad inspiration” only, and the results are pretty generic when you try.

It’s slower than competitors. Not dramatically, but you’ll notice the wait compared to Suno or Udio.

And here’s the nuanced take: the music community is genuinely split on quality. The audio fidelity — the way instruments are mixed, the spatial depth, the production quality — is excellent. A 15-year audio engineer called it superior to both Suno and Udio on pure sound design. But the musical creativity — the melodic choices, the lyrical ideas, the arrangements — is where opinions diverge. Some Japanese musicians called the songwriting “two-years-ago level.” Others find that sloppy prompts produce mediocre results, but detailed structural prompts produce genuinely impressive output.

Fair assessment: Lyria 3 Pro sounds better than its competitors but doesn’t always write as interestingly. Your prompt quality matters more here than with any other AI music tool.

Lyria 3 Pro vs Suno vs Udio: How They Compare

This is the comparison everyone’s asking about. Here’s the honest breakdown:

Feature	Lyria 3 Pro	Suno v5	Udio v4
Max track length	3 min	~5 min	Full songs
Free tier	No	Yes (50 credits/day)	Yes (100 credits/mo)
Stem export	No	Yes	Yes
Inpainting	No	No	Yes
Song structure control	Yes	Limited	Limited
Vocal quality	Good	Best	Great
Cover art generation	Yes	No	No
API access	Yes	Limited	Limited
Commercial use	Yes	Yes (paid plans)	Yes (paid plans)
SynthID watermark	Yes	No	No

Choose Lyria 3 Pro if: You’re already paying for Gemini, you need API access for an app or service, you want the legal safety of licensed training data, or you’re creating content within Google’s ecosystem (YouTube, Google Vids).

Choose Suno if: You want the best vocal quality, need longer tracks, want stem export, or you’re on a budget (free tier is generous). Suno has roughly 950 million users and a ~$2B valuation for a reason — the output quality is still the benchmark.

Choose Udio if: You want to fine-tune sections of your track after generation. Udio’s inpainting feature is genuinely unique — regenerate just the bridge, just the intro, just 10 seconds in the middle. For producers who want control, this is huge.

The Copyright Angle (Worth Knowing)

One advantage Lyria 3 Pro has that doesn’t show up in feature comparisons: Google says it trained on licensed data.

Both Suno and Udio faced copyright lawsuits from major record labels in 2025 before reaching settlements. Google’s approach — using “materials that YouTube and Google has a right to use under our terms of service” — shields it from similar legal challenges. Universal Music Group even negotiated specific AI guardrails with YouTube.

Does this matter if you’re making background music for a YouTube video? Probably not today. But if you’re building a product, licensing music for commercial use, or in any situation where legal clarity matters — it’s worth noting. Google confirmed commercial use is allowed for paid subscribers.

That said, indie artists did sue Google in early March 2026 over YouTube training data. So “licensed” doesn’t mean “uncontested.”

5 Prompts That Actually Work Well

Here are prompts that consistently produce good results, based on community testing:

1. The Content Creator Background Track

“Create a 2-minute upbeat lo-fi hip hop instrumental with jazzy piano chords, a mellow drum beat, vinyl crackle, and soft bass. Tempo around 85 BPM. No vocals. Warm and cozy feeling, perfect for study or work videos.”

2. The Podcast Intro

“Create a 30-second electronic intro jingle. Start energetic with a synth hook, build for 20 seconds, then cleanly resolve. Modern and professional, similar to tech podcast intros. No vocals, 128 BPM.”

3. The Storytelling Song

“Create a 2-minute indie folk song with male vocals. Acoustic guitar and light percussion. Lyrics about leaving a small town to chase a dream. Start quiet with just guitar, build to a full arrangement in the chorus, end stripped back. Key of G major.”

4. The Cinematic Score

“Create a 3-minute orchestral piece. Start with a solo cello playing a melancholic theme. Gradually add strings, then a full orchestra builds to an emotional climax at 2 minutes. Resolve gently with the cello theme returning alone. Film score style, dramatic.”

5. The Genre Experiment

“Create a 1980s Japanese city pop track with funky bass, bright electric piano, lush synth pads, and upbeat female vocals in Japanese. Tempo 110 BPM, the feel should be driving at sunset along the coast. Include a saxophone solo in the bridge.”

Who Should Use Lyria 3 Pro

It’s a good fit if you:

Already pay for Gemini (no extra cost)
Need background music for content — videos, podcasts, social media
Want API access to build music into an app or workflow
Care about legal clarity around training data
Want quick results without learning a new platform

It’s not the right tool if you:

Want free AI music generation (try Suno’s free tier)
Need to edit or remix generated tracks (try Udio)
Want to separate vocals from instrumentals (try Suno)
Need songs longer than 3 minutes
Want the absolute best vocal quality (Suno still leads here)

The Bottom Line

Lyria 3 Pro isn’t the best AI music generator on every metric. Suno makes better-sounding songs. Udio gives you more control. Both have free tiers.

But Lyria 3 Pro does something neither competitor can match: it lives inside an ecosystem. Gemini on your phone, Google Vids for presentations, Vertex AI for developers, YouTube for distribution. If you’re already in Google’s world, adding AI music is just one prompt away.

And with OpenAI shutting down Sora and pulling back from creative tools, Google’s timing is sharp. They’re betting big on creative AI while others are retreating. Whether Lyria 3 Pro earns its place in your workflow depends on what you need — but it’s worth the 2 minutes to try.

Sources: