Lyria 3.5 Guide: How to Use Google’s AI Music Generator

Quick Answer

Lyria 3.5 is Google DeepMind’s latest AI music generation model. You describe the music you want—or add images—and it generates a high-fidelity stereo track with instruments, optional vocals, and lyrics.

The basic loop is simple:

• Describe a song with a text prompt, and optionally add images.
• Specify details such as genre, mood, vocals, tempo, structure, and length.
• Generate the track.
• Listen for weak sections or mismatched details.
• Change one or two instructions and generate another version.

It is easy to start, but matching a precise idea takes iteration. Compared with earlier Lyria models, Lyria 3.5 stands out for richer musicality, stronger song structure, better lyrics, more expressive vocals, improved pronunciation, and easier tempo and duration control.

What Is Lyria 3.5?

Lyria 3.5 is Google’s flagship music generation model. It is optimized for full-length songs with verses, choruses, and bridges, rather than only short loops.

Official capabilities include:

• Text-to-music and image-to-music generation
• High-quality 44.1 kHz stereo audio
• Tracks up to about 3 minutes
• Vocal or instrumental output
• Lyrics with better prompt adherence and structural awareness
• Tempo and duration control
• Vocals in multiple languages

Instead of composing every note yourself, you act like a creative director: describe the finished track, then refine the result through new generations.

Useful things you can specify include:

• Genre or a clear genre blend
• Mood and energy
• Instruments
• Vocal or instrumental output
• Vocal character and language
• Tempo or BPM
• Song duration
• Intro, verse, chorus, bridge, and outro structure
• Production style
• Subject or theme for lyrics

How To Use Lyria 3.5

There are no lives, scores, or win conditions. The workflow is: describe, generate, listen, refine.

1. Start With the Musical Identity

Begin with genre, mood, and energy rather than writing a huge prompt immediately.

Example:

Dreamy synth-pop with a nostalgic late-night mood, warm analog synths, female vocals, and a slow-building chorus.

That already gives the model several useful anchors without overloading it.

2. Add Tempo and Instrumentation

If the first output feels too random, become more specific.

Try adding:

• 92 BPM
• Funk bass
• Clean electric guitar
• Analog synthesizers
• Punchy electronic drums
• Soft piano intro

Tempo and duration are especially useful when the song needs to fit a video, ad, or exact runtime.

3. Decide Between Vocals and Instrumental

For a vocal track, describe both the musical style and how the singer should perform.

Useful descriptors include:

• Intimate
• Powerful
• Airy
• Melancholic
• Energetic
• Conversational
• Dramatic

For background music, explicitly request an instrumental track with no vocals so the model does not add singing.

4. Describe the Song Structure

This matters more for longer songs.

A prompt can request:

• Short atmospheric intro
• Verse
• Pre-chorus build
• Large melodic chorus
• Second verse
• Bridge with reduced instrumentation
• Final chorus
• Clean ending

In more technical workflows, section tags such as [Intro], [Verse], [Chorus], [Bridge], and [Outro]—or timestamps like [0:00-0:12]—can help define when the song should change.

5. Generate and Judge the Result

Do not judge only the first 10 seconds. Listen for:

• Does the hook actually develop?
• Are the vocals understandable?
• Does the chorus feel different from the verse?
• Does the requested instrument appear prominently enough?
• Does the song keep its style?
• Does the ending feel intentional?

Treat each generation as another attempt. Lyria generations are typically single-turn, so you refine by rewriting the prompt rather than editing the finished audio note by note.

Where Can You Try Lyria 3.5?

As of September 2026, Lyria 3.5 is available in several Google products:

• Gemini app on web and mobile: the easiest starting point for most people. You can describe a genre, choose vocal or instrumental, use templates, and pick short or longer tracks.
• Google Flow Music: a creator-focused music workspace, and the first place Lyria 3.5 launched.
• Google AI Studio and the Gemini API: for developers and more technical experiments. The model ID is lyria-3.5.
• Google Vids: for adding custom music to video projects.

For casual use, start in Gemini. For more control over longer songs, Flow Music or the API is usually better.

How Difficult Is Lyria 3.5?

Starting difficulty: Easy

Anyone who can describe a song can generate something quickly.

Mastery difficulty: Medium

The challenge is controlling several musical variables without creating contradictory instructions.

You will usually get better results by progressing from:

Genre → mood → instrumentation → vocals → tempo → structure → duration

rather than dumping every requirement into the first prompt.

Tips for Better Lyria 3.5 Music

Use Musical Details Instead of Vague Adjectives

"Make an amazing song" gives Lyria very little direction.

A better prompt would say:

Upbeat indie pop at 118 BPM with jangly guitars, live-sounding drums, bright bass, youthful vocals, and a catchy chorus that arrives within 40 seconds.

Specific musical information gives the model something it can act on.

Combine Genres Carefully

Genre mixing is one of the most entertaining ways to use Lyria, but avoid stacking five unrelated styles.

Try combinations with a clear hierarchy:

• Jazz-inspired hip-hop
• Bossa nova with trip-hop production
• Funk-pop with electronic elements
• Cinematic orchestral music with ambient techno textures

Make one genre the foundation and use the second as an influence.

Specify BPM When Rhythm Matters

Words such as "fast" and "slow" are subjective.

If groove matters, try an actual tempo:

• 70–85 BPM: relaxed or heavy
• 90–110 BPM: moderate groove
• 115–130 BPM: energetic dance territory
• 140+ BPM: faster, more intense styles

You do not always need BPM, but it helps when generations repeatedly feel too fast or too slow.

Describe the Vocal Performance

Do not stop at "female vocal" or "male vocal."

Add performance information such as:

soft intimate verses with a more powerful, emotional chorus.

Details about emotion, intensity, language, and delivery can be just as important as choosing a vocal type.

Tell It What Should Change Over Time

A good full song should evolve.

Instead of requesting the same sound for the entire track, try instructions such as:

• Begin minimal and gradually add instruments.
• Make the second chorus bigger than the first.
• Drop the drums during the bridge.
• Finish with a short instrumental outro.

This helps prevent the result from sounding like one idea repeated for several minutes.

Use Images as Inspiration

Lyria 3.5 can use images as creative input. In the API, you can provide multiple images alongside a text prompt.

This works especially well for:

• Travel photos
• Fantasy artwork
• Game screenshots
• Landscape photography
• Character art
• Album-cover concepts

Use the image as the mood anchor, then add a short instruction describing how you want it interpreted musically.

What Makes Lyria 3.5 Stand Out?

The biggest improvement is not only cleaner audio. It is better control over a complete musical idea.

Google highlights these upgrades:

• Richer, more natural melodic structure
• Higher-quality lyrics with stronger prompt adherence
• More expressive, emotionally nuanced vocals and improved pronunciation
• Easier control of tempo and duration
• Better suitability for complete songs instead of short loops

That makes Lyria 3.5 especially useful if earlier AI music tools sounded impressive for 20 seconds but became repetitive across a longer track.

What Are the Limitations?

Lyria 3.5 can still misunderstand a complicated prompt or produce a track that technically follows the instructions but does not match the exact idea you imagined.

Common reasons to regenerate include:

• Weak or forgettable chorus
• Instrument balance that differs from the prompt
• Lyrics that feel generic
• A vocal performance that does not match the requested emotion
• Sections that transition too abruptly
• Genre combinations that do not blend naturally

Other limits to keep in mind:

• Generation is usually single-turn. You generally cannot surgically edit one section after the track is created.
• Prompts that ask to clone a specific artist’s voice or reproduce copyrighted lyrics are blocked.
• Results can vary even with the same prompt.
• Every generated track includes an imperceptible SynthID watermark so the audio can be identified as AI-generated.

The best approach is usually to change one or two important instructions between generations instead of rewriting the entire prompt every time.

Who Is Lyria 3.5 For?

Lyria 3.5 is worth trying if you enjoy:

• Experimenting with unusual genre combinations
• Creating custom songs from ideas or jokes
• Making background music for personal or video projects
• Turning images into soundtracks
• Exploring different vocal styles and languages
• Iterating on song concepts without traditional production skills

It is less compelling if you expect deterministic control over every note, chord, lyric, and mix decision. Traditional music software still gives much finer manual control.

Final Take

Lyria 3.5 turns AI music generation into an accessible creative loop: describe, generate, listen, refine, and repeat. Its biggest appeal is moving beyond short musical experiments toward more structured tracks with better vocals, lyrics, tempo control, and evolving song sections.

For beginners, start in the Gemini app with a simple musical identity. Add specific controls only where necessary, then iterate until the structure, vocals, and mood line up with what you want.

If you enjoy experimenting with AI-generated music, Lyria 3.5 is worth trying. The key is not writing the longest possible prompt. Start clear, add constraints carefully, and regenerate until the track feels intentional.