You're used to typing prompts into generative AI tools like ChatGPT, Gemini or Grok to get answers, images and even videos. With Lyria 3, Google wants you to do the same thing for music. You can type a prompt, upload audio or an image, and the model can turn that idea into a song in seconds. 

Lyria 3 launched in February as Google's most advanced music generation model. In March, Google introduced Lyria 3 Pro, an even more advanced music model. 

Competitors like Suno and Udio have dominated the viral AI song space. Now, Google wants a piece of that pie, too. The company has experimented with music models before, but Lyria 3 and Lyria 3 Pro are designed to create longer songs with better sound quality and more structure.

Let's dive deeper.

What is Lyria 3?

Lyria 3 is Google DeepMind's latest multimodal music AI model. This means it can work with more than one kind of input, including text, audio and images, and it composes a soundtrack that fits the mood and timing of the visuals.

The tool can generate high-fidelity music with more natural flow from note to note, and it can now keep melody, rhythm and style more consistent from the beginning of a track to the end. 

Lyria 3 in Gemini is included in all of Google's AI plans, with higher limits for paid subscribers and for Lyria 3 Pro. According to the company, Lyria 3 Pro is available across products, including Vertex AI, Google AI Studio, the Gemini API, Google Vids, the Gemini app and ProducerAI, a music creation platform that Google brought into Google Labs in February. 

ProducerAI is pitched less as a one-shot song generator and more as a creative partner. You can generate full songs, but also specific elements, like beats, melodies or hooks, which you can keep shaping.

Enlarge Image

Getting started with Lyria 

Lyria 3 in Gemini is fairly straightforward. You describe the kind of song you want, or upload an image, and Gemini turns that into a 30-second track with AI-generated cover art. If you choose the Thinking model, it can generate longer tracks, like the one I created.

Enlarge Image

In Google's developer tools and cloud products, the Pro model can generate tracks that better handle song structure, including intros, verses, choruses, bridges and outros. In the Gemini API, Lyria 3 clip models generate 30-second, 48kHz stereo audio.

Both models offer deep control over the musicality, allowing you to specify complex genres, instrumentation and even the era of the sound, like "'70s analog warmth."

Enlarge Image

Lyria 3 Pro supports multiturn editing with some limitations, Google says, meaning you can keep iterating in a conversation instead of starting over every time. But if the model heads in the wrong direction, it's usually easier to just start over than to keep refining it.

Mastering the prompts 

Prompting in Lyria 3 is where the creative magic happens. A simple list of keywords will generate a song, but to control the models, Google suggests you use this framework: [Genre and style] + [Mood] + [Instrumentation] + [Tempo and rhythm] + [Vocal style & language] + [Lyrics].

You can check out Google's prompting guide for more detailed instructions and ideas. You can also ask a chatbot to suggest a prompt for you.

"Words don't translate to music one-on-one," John von Seggern, CEO of Futureproof Music School, tells me. "You can't type 'make me a catchy pop song' because that'll always end up as something else." 

He thinks people are more open to "using AI to make little building blocks … as opposed to just using words to shape an entire composition."

Lyrics and language

Lyria 3 currently supports vocal and lyric generation in eight languages: English, Spanish, French, German, Portuguese, Hindi, Japanese and Korean. But it's not a basic robotic voice like you're used to hearing narrate clips on socials. The AI tool actually produces audio with emotional inflection that matches the genre you've requested.

There are two main ways to handle the lyrics for your tracks. You can describe a specific theme or mood and let the tool write and sing the words for you, or you can provide your own custom lyrics in your prompt. You can also time-align your lyrics and control exactly when the lyrics start and end within the track.

Lyria 3 manages translations and vocal styles. If you write your instructions in English but specify that you want the actual song to be performed in another supported language, the tool will handle the translation and the vocal delivery for your prompt to, for example, "make this a K-pop track in Korean."

Who should use Lyria?

Google is trying to position Lyria 3 as a flexible creative tool, not only a toy for hobbyists. The most obvious use cases are for digital creators and social media influencers. Think background music for short videos, podcasts, pitches, demos and social posts. 

"I felt it was designed for like, 'I'm making a three-minute video for YouTube, and I need some kind of average backing music under my YouTube documentary.' It'd be perfect for that," von Seggern says.

But don't think you're going to become the next big star by prompting your way to fame just yet. 

"If you want more granular control over what you're doing, then you're going to need to know what you're doing on some level," he adds.

Google is also suggesting it as a tool for experimentation and education, especially for students who may not have access to instruments or a well-funded music program. 

Is the music actually good?

The internet is already clogged with AI slop. The last thing we need is another flood of horrible content. Music may be a tougher category to master than text or images because people form emotional bonds with songs. That makes tools like Lyria 3 interesting, but also more sensitive to scrutiny.

Von Seggern thinks the audio quality is "quite good," but he was less impressed by its originality. "I was disappointed that it seemed pretty generic," he said.

What about copyright and ownership?

The US Copyright Office's January 2025 report states that AI-generated work can be copyrighted only when a human adds meaningful creative input, but not when the work comes from prompts alone.

"What that is is sort of up for interpretation. But it still needs to be some human creation," Relani Belous, founder of Belous Law Corporation and The Trademark Channel, tells me. 

Before you start planning how to monetize your AI-generated tracks, Belous says people should think twice before hitting publish. She recommends understanding the terms and the legal risk, and consulting a legal expert. (Belous adds the disclaimer that this is a legal opinion, not legal advice.)

"When you have any sort of disruptive technology, there's going to be lawsuits," Belous says, adding that in lawsuits, people always go after the "deepest pocket." 

And the lawsuits have already started. 

A group of independent musicians and songwriters sued Google in March, accusing the company of training Lyria 3 on copyrighted recordings from "at least 44 million clips and 280,000 hours of music" pulled from YouTube without permission or payment.

Google says it trained Lyria 3 models "using materials that YouTube and Google have a right to use under terms of service, partner agreements and applicable law."

Creatives often approach their work from an artistic mindset rather than a business one, which can leave them more vulnerable when contracts, management and money come into play, Belous says. 

"There is a historical context, especially with musicians not getting what they should have," Belous says.

Google says music generated in Gemini is embedded with SynthID. This is an invisible, inaudible watermarking technology embedded directly into AI-generated audio content. The company also says Lyria 3 Pro is designed to avoid mimicking existing artists. 

"You can't copyright a style ... like a cowboy movie set on a ranch. But when you get down to the melody and distinctive elements, that's where people bring in musicologists," Belous says. 

She also says that fair use, which is part of copyright law, is widely misunderstood and isn't just an excuse to infringe. "Fair use is not a right. Fair use is an affirmative defense, which means you've done something that is infringing, but it means you have an affirmative defense to do that."

What's next for AI-generated music?

Google says it believes in working alongside artists to enhance human creativity, not to replace it.

Grammy-winning artist Wyclef Jean used Lyria 3 as a creative tool during the development of his song, Back From Abu Dhabi. Jean says, "There's one thing we have over AI, and that's soul. There's one thing that AI has over us, and that's infinite creation," calling the combination of both "invincible."

And it seems other artists, like Grimes, wouldn't mind if their music is used for training or fusion with AI, but for compensation.

Others have seemed to try to take it to the dark side. In 2025, Spotify removed an AI-generated song that had been uploaded under the profile of late country singer Blaze Foley, who was killed in 1989. 

"I think there's an enormous amount of uncertainty and risk here that needs to be balanced out, and that's something that will be vetted out over the years, like we see with new technology," Belous says. She also noticed that licensing agreements are changing and AI-related clauses are being added to contracts.

Lyria 3 is technically impressive and seems useful in some cases. But music is personal. If AI music is going to stick, it will need to be more than soulless background noise.