Lyria 3.5

Name: Lyria 3.5
Author: Google DeepMind

byGoogle DeepMind

Full-length studio musical arrangements and expressive vocal performances guided by structural timestamps and visual inspiration

Compose full song

How Lyria 3.5 works

Direct tempo, structural timestamps, and visual moodboards to generate complete stereo arrangements in three focused actions.

Fuser Text node with a folk ballad prompt naming key, tempo and song sections, plus a misty lake photo, wired into Lyria 3.5.

Define Structure and Style

Set the key, tempo in BPM, and section markers like [Intro], [Verse], and [Chorus] alongside any desired lyrics.

Lyria 3.5 node in Fuser with the misty lake photo on its Image Inspiration input and a 2:37 track ready to play.

Attach Visual Inspiration

Provide a concept still or moodboard to align the harmonic texture and emotional palette with visual aesthetics.

Lyria 3.5 node in Fuser with its finished 2:37 folk ballad ready to play and a Lyrics output beside the audio.

Generate Full Master and Lyrics

Produce a cohesive stereo arrangement up to three minutes long alongside synchronized lyrics text.

What Lyria 3.5 is good at

From long-form architectural planning to image-guided scoring, discover what sets this music generation model apart.

Timestamped Multi-Section Song Planning

Orchestrate full compositions up to 3 minutes with designated intros, verses, choruses, and outros using bracketed timestamp markers and explicit BPM settings.

Expressive Vocal Performance and Lyrics

Synthesize natural human vocal delivery with expressive breath dynamics and articulated diction, generating synchronized lyrics alongside the stereo mix.

Image-Conditioned Soundtrack Generation

Condition your musical direction with visual artwork, film stills, or moodboards to guide genre, instrumentation, and emotional resonance.

Studio-Grade Stereo Soundstage

Deliver balanced studio-quality mixes featuring distinct low-end definition, clear mid-range instruments, and SynthID imperceptible watermarking.

Made with Lyria 3.5

Explore full-length musical arrangements spanning cinematic scores, dance grooves, and vocal ballads generated from text and image prompts.

Indie dream pop arrangement with female vocals

Cinematic sci-fi orchestral crescendo

Deep melodic techno club track

Acoustic folk ballad with fingerstyle guitar

Lo-fi chillhop beat with electric piano

What people build with Lyria 3.5

See how music producers, filmmakers, brand directors, and game developers build soundtrack-ready audio with Lyria 3.5.

Film and Game Soundtrack Scoring

Score original multi-section cues that shift dynamics alongside scene cuts using timestamped prompts and concept art.

Songwriting and Demo Prototyping

Draft complete songs with custom verse-chorus lyrics, tempo locks, and genre-authentic vocal performances.

Commercial Brand Scoring

Align background music seamlessly with campaign visuals by feeding keyframe stills directly into the generation prompt.

Content Creator Background Music

Produce copyright-cleared, full-length background songs tailored to specific narrative pacing and video moods.

Genre Exploration and Stylistic Studies

Synthesize period-accurate timbres from 70s soul to 90s shoegaze with authentic instrument layering and vintage tape saturation.

Lyria 3.5 specs

Plans and credits

Inputs: Prompt (text), Image Inspiration (image)
Output: audio, text
Cost: 138 credits per run

Frequently Asked Questions

Lyria 3.5 is engineered for full-length structured songs up to 3 minutes, complete with multi-section arrangements (intro, verse, chorus, bridge, outro), realistic vocal singing, and multi-modal image conditioning. It plans harmonic progressions and melodic motifs across the entire piece rather than repeating short loops.

Use timestamp markers and bracketed structural tags in your prompt, such as [00:00 - 00:20] [Intro] and [00:20 - 00:50] [Verse]. To enforce specific singing, prefix lines with Lyrics: inside the section blocks, and define the key signature and BPM upfront to anchor the groove.

Lyria 3.5 generates a unified, mixed stereo audio file (and transcribed lyrics) rather than isolated multitrack stems (such as separate drum, bass, or vocal tracks). It is also not suited for non-musical Foley sound effects or sub-millisecond audio editing.

Lyria 3.5 was developed by Google DeepMind. The architecture was initially introduced in November 2023, bringing long-form harmonic planning and SynthID imperceptible audio watermarking to generative music synthesis.