Google · From prompt to full song, complete with vocals.

Lyria 3.5

Studio-Grade AI Vocals, Fully Controllable

Lyria 3.5 is Google DeepMind's latest music model, turning a text prompt or an image into a full song with expressive vocals, cleaner lyrics, and hands-on control over tempo and length. It launched in Google Flow Music on July 29, 2026.

3 min
Max song length
July 29, 2026
Release date
SynthID
Watermarked output
Text + Image
Prompt inputs
What's new

Studio-Grade AI Vocals, Fully Controllable

01

More expressive vocals

Sings with more realistic emotion and noticeably better pronunciation than Lyria 3, closing the gap with vocal-focused rivals.

02

Sharper lyric generation

Higher-quality lyrics with stronger prompt adherence and real awareness of song structure like verses and choruses.

03

Text or image to music

Start from a written prompt or upload an image and let Lyria 3.5 translate its mood into an original, high-fidelity track.

04

Direct tempo and length control

Dial in the tempo and duration of the output instead of regenerating, for faster iteration toward the take you want.

Overview

Model at a glance

Lyria 3.5 is Google DeepMind's music generation model, released on July 29, 2026 inside the Google Flow Music app. It is the third notable step in the Lyria 3 line within roughly five months: Lyria 3 arrived in February 2026 generating short 30-second clips from text or images, Lyria 3 Pro followed in late March extending tracks to about three minutes with section-level structure, and Lyria 3.5 now refines output quality across the board. Rather than chasing a single new headline feature, the 3.5 update concentrates on the parts of a song listeners judge most harshly — the singing, the words, and the overall musical feel.

The model takes a natural-language description, or an uploaded image, and produces a complete track with vocals and instrumentation. Google positions Lyria for commercial creators: the Lyria 3 line is described as trained on licensed data, and Flow Music's paid tiers advertise commercial-use rights, which sets it apart from several rivals whose training practices remain contested in court. Every clip Lyria produces is embedded with SynthID, Google's imperceptible, machine-detectable watermark, so AI-generated audio can be identified downstream. At launch Lyria 3.5 is available only through Google Flow Music at no extra cost to users; the earlier Lyria 3 and Lyria 3 Pro models are offered separately on Vertex AI in preview.

Model
Lyria 3.5
Developer
Google DeepMind
Modality
Text/image to music with vocals
Released
July 29, 2026
Max duration
Up to ~3 minutes
Watermark
SynthID on all output
Availability
Google Flow Music
Commercial use
Via Flow Music paid tiers
Generational leap

FeatureLyria 3Lyria 3.5NEW
February 2026July 29, 2026
~30-second clipsUp to ~3 minutes
Baseline vocalsMore expressive, better pronunciation
Basic lyric generationHigher quality, structure-aware
LimitedDirect tempo and duration control
Text or imageText or image

Lyria 3.5 is coming soon to SharkFoto

Google DeepMind's expressive AI music model is on our roadmap. Until it lands, dive into SharkFoto's available AI tools to create and edit your visuals — and check back for Lyria 3.5's arrival.

Try Lyria 3.5 now