Google's Lyria 3 Pro generates full songs up to three minutes long

Google DeepMind has released Lyria 3 Pro, a music generation model that creates complete tracks up to three minutes long, a sixfold increase over the 30-second clips produced by the standard Lyria 3 model that launched in February.^[1][2] The new model is rolling out starting today across the Gemini app, Vertex AI, Google AI Studio, Google Vids and ProducerAI.^[2]

What it does

Lyria 3 Pro's headline capability is structural awareness. Where earlier AI music generators often struggled with long-range coherence across a full track, Lyria 3 Pro is designed to understand the architecture of a song. Users can prompt for specific elements like intros, verses, choruses and bridges, and the model builds transitions between them.^[2] Google describes the result as tracks with "professional-grade structural awareness" that maintain musical consistency across the full duration.^[3]

The model also supports granular control through natural language. Users can set specific tempos, direct when lyrics start and end within a track, and define vocal styles and acoustic preferences.^[2][3] Beyond text prompts, the Lyria 3 family accepts image inputs that influence the mood and atmosphere of the generated audio.^[3]

In our testing, we tried both ends of the prompting spectrum. Uploading a photo of my bengal cat lounging on a bed produced a hazy bedroom pop track titled “King of the Cotton”. The model analyzed the image, inferred a mood and generated a finished composition with cover art in seconds.

At the other extreme, a detailed text prompt specifying a dramatic male baritone over a slow orchestral arrangement at 70 BPM, with named instruments, a verse-to-chorus dynamic arc and specific mixing directions for reverb and compression, returned a track called "The Horizon Is A Blade" that followed the brief closely.

The two demos illustrate the range Lyria 3 Pro is targeting: casual image-to-music generation for everyday users and production-level text control for creators who want to direct every element.

For developers, Google is offering two distinct API variants. Lyria 3 Pro (model ID: lyria-3-pro-preview) handles full-length song generation up to approximately three minutes. A companion model called Lyria 3 Clip (lyria-3-clip-preview) is optimized for speed and generates 30-second clips, intended for rapid prototyping and social media assets [3]. Both are available in public preview through the Gemini API.

Where it's available

Google is pushing Lyria across a broad surface area:

• Gemini app: Lyria 3 Pro is available to Google AI Plus, Pro and Ultra subscribers for longer generations with enhanced customization.^[2]

• Vertex AI: Now in public preview for businesses requiring on-demand audio at scale, targeting use cases from gaming soundtracks to creative tool integration.^[2]

• Google AI Studio: Both models are available alongside Lyria RealTime in a new dedicated music generation workspace with two creation modes. Text mode accepts natural language descriptions; Composer mode lets users build songs section by section, setting timing and intensity for each part individually.^[3]

• Google Vids: Rolling out this week to Google Workspace customers and Google AI Pro and Ultra subscribers for adding custom music to video projects.^[2]

• ProducerAI: The recently acquired music creation platform, formerly known as Riffusion [9], now uses Lyria 3 Pro to offer what Google describes as an agentic experience for iterating on comprehensive songs. It is available globally with free and paid plans.^[2][4]

Google has also built demo applications in AI Studio showcasing potential integrations, including a tool that analyzes uploaded video with Gemini 3 Flash and generates a matching soundtrack, and an alarm clock app that creates a daily song incorporating weather, location and calendar data.^[3]

Competitive landscape

The release lands in the middle of a rapidly consolidating AI music market. Suno, the current market leader, reported reaching 2 million paid subscribers and $300 million in annual recurring revenue in February 2026, up from $200 million just three months earlier.^[5] The company is valued at $2.45 billion following a $250 million Series C [5]. Udio, founded by former DeepMind researchers, has been reshaping its business around label licensing deals that restrict its generated tracks to a walled garden model.^[6]

Both Suno and Udio have been settling copyright infringement lawsuits filed by major record labels in 2024. Warner Music Group settled with both companies. Universal Music Group settled with Udio under terms that required the startup to pivot toward a fan engagement platform rather than open-ended generation. Sony has not settled with either.^[6]

Google has taken a different approach to the licensing question. The company states that Lyria 3 was trained using materials that YouTube and Google have the right to use under existing terms of service, partner agreements and applicable law.^[2] The Lyria 3 model card, which covers Pro and subsequent versions, indicates the underlying architecture uses latent diffusion applied to temporal audio latents, trained on audio datasets annotated with text captions at varying levels of detail.^[7]

Google also acquired ProducerAI in February 2026 ^[4], roughly a week after the original Lyria 3 launch.^[1] ProducerAI was originally known as Riffusion, an open-source text-to-music project that went viral in December 2022.^[9] The startup had raised a $4 million seed round in 2023, with The Chainsmokers among its backers.^[9] Following the acquisition, ProducerAI's team joined Google Labs and Google DeepMind.^[4][9]

Safeguards and creative partnerships

All Lyria 3 and Lyria 3 Pro outputs are embedded with SynthID, Google's imperceptible watermarking system for identifying AI-generated content.^[2] The model is designed not to mimic named artists. If a prompt includes an artist's name, the model treats it as broad inspiration rather than attempting to replicate their style. Google also employs filters to check outputs against existing content.^[2]

On the creative partnership side, Google has been working with musicians through its Music AI Sandbox program. Grammy-winning producer Yung Spielburg used Lyria in the scoring process for the Google DeepMind short film "Dear Upstairs Neighbors". DJ and producer François K used the model in an iterative process to create a forthcoming release. Wyclef Jean has also collaborated with Google on music through the program.^[4]

What to watch

Lyria 3 Pro represents Google's bid to compete with dedicated AI music startups while leveraging its unique advantages: a massive existing user base through Gemini, enterprise distribution through Vertex AI, and training data access through YouTube partnerships. At three minutes, Lyria 3 Pro's maximum track length still trails Suno, whose v5 model can generate up to eight minutes in a single pass.^[10] Google's pitch rests more on structural awareness and integration breadth than on raw duration, though the quality comparison will depend on real-world use.

The broader question remains whether AI-generated music will complement or compete with human creators. Deezer has estimated that 50,000 fully AI-generated songs are delivered to its platform daily, and the streaming service claims 97% of listeners cannot distinguish them from human-made tracks.^[6] Spotify has responded with strengthened policies targeting spam, impersonation and streaming fraud related to AI-generated music.^[6] iHeartRadio has gone further, launching a "Guaranteed Human" program pledging not to use AI-generated personalities or play AI music featuring synthetic vocalists posing as human.^[6]

For now, Google is positioning Lyria as a creative tool rather than a replacement for human musicianship. Whether the market, the industry and the listening public agree will shape how this technology evolves.