Google DeepMind's music model: sound, timed sections and lyrics in one prompt, plus an optional mood image. Songs or instrumentals up to about three minutes, 35 credits each.

Example Image
Google DeepMind's music model: songs or instrumentals up to ~3 minutes from one prompt, with timed sections and lyrics.

Generator examples

Real Lyria 3.5 outputs · generated on Voor, October 4, 2026

Lyria 3.5, heard line by line

Lyria 3.5 is Google DeepMind's music model. One prompt holds the sound, a timed section plan and your lyrics. These unedited Lyria 3.5 tracks show when each line was actually sung — tap a line to check it.

0:00 / 2:07
Style
Indie synth-pop
Voice
Breathy female lead, stacked chorus harmonies
Asked for
About 2 minutes
Delivered
2:07

Lyric sheet · measured from the output · tap a line to hear it

  1. Intro0:00

    Pads and guitar arpeggio, no vocal

  2. Verse 10:14
  3. Chorus0:45
  4. Verse 21:04
  5. Final chorus1:31

    Same four chorus lines, sung once

  6. Outro1:48

    Vocal ends early; the guitar outro runs about 20 seconds

Every written line was sung in order. The vocal entered two seconds after the planned 0:12 and the song finished at 2:07 against a two-minute request.

Timestamps in, timing out

Lyria 3.5 follows a timed plan

No length slider: write “about 2 minutes” or mark sections such as [0:42-1:05] Chorus. Most Lyria 3.5 sections landed within three seconds of the plan.

Window Seat · measured from the sung lyrics

SectionAskedHeardDrift
Verse 10:120:14+2.4s
Chorus0:420:45+2.5s
Verse 21:051:04−1.0s
Final chorus1:301:31+0.7s
Last sung wordChorus ran short; the outro filled the gap1:551:48−7.5s
End2:052:07+1.9s

Canyon Sunrise · measured from loudness changes

SectionAskedHeardDrift
Build0:250:26+1.0s
Full orchestra0:550:53−2.0s
Resolve1:201:22+2.0s
End1:301:31+0.6s
A hiker with a backpack stands on a narrow trail between misty canyon walls as the sun rises
Mood image attached to the Canyon Sunrise prompt

Optional image input

Score a picture with Lyria 3.5

Add a picture in the Reference images field and Lyria 3.5 uses it as a mood guide. Our prompt also described the scene, so keep genre, length and structure in the text. For a moving shot, make the clip with image to video and score the same frame.

The exact Window Seat prompt

Write a Lyria 3.5 prompt in three layers

  1. 01 · Sound

    Dreamy indie synth-pop, 104 BPM, A major. Warm analog pads, chorus-soaked clean guitar arpeggios, punchy drum machine, round synth bass. Soft, breathy female lead vocal with stacked harmonies on the chorus. Wistful but hopeful, like a night train crossing a lit-up city. About 2 minutes long.

  2. 02 · Plan
    • [0:00-0:12] Intro: pads and guitar arpeggio only.
    • [0:12-0:42] Verse 1: sparse drums, close vocal.
    • [0:42-1:05] Chorus: full drums, harmonies open up.
    • [1:05-1:30] Verse 2.
    • [1:30-1:55] Final chorus, biggest moment.
    • [1:55-2:05] Outro: guitar arpeggio fades out.
  3. 03 · Lyrics

    Lyrics: then [Verse 1], [Chorus], [Verse 2], [Chorus] — the words in the sheet above. Repeat a chorus in full where it should be sung again.

Window Seat

117/117 written words sung

No extra words added

Mercado de Domingo

111/114 written words sung

+22 words of ad-lib outro the model added

Checked against a speech-recognition transcript. End the plan with a timed outro to stop Lyria 3.5 filling spare seconds with ad-libs.

Settings, credits and alternatives

Lyria 3.5 on Voor

Inputs
Prompt up to 5,000 characters; optional reference image.
Vocals
English, Spanish, German, French, Portuguese, Hindi, Japanese, Korean, or instrumental.
Output
One MP3 up to about three minutes, 44.1 kHz stereo, SynthID watermark.
Credits
35 per track at any length. New accounts get free welcome credits.
Not offered
Duration number, seed, negative prompt, reference audio, stems, or imitation of a named artist.
  • MiniMax Music 3

    Switch when you need a separate lyrics field and a duration number up to five minutes.

  • Mureka AI

    Switch when you need a fixed female or male lead, or a song written from a one-line idea.

  • AI jingle maker

    Switch when you need a short sung hook with a clean button ending.

  • Text to speech

    Switch when words must be spoken, then lay a Lyria 3.5 instrumental under them.

Lyria 3.5 FAQ

What is Lyria 3.5?

Lyria 3.5 is Google DeepMind's music generation model. One text prompt returns a complete vocal song or an instrumental of up to about three minutes as a 44.1 kHz stereo MP3.

Can Lyria 3.5 sing my own lyrics?

Yes. Add a Lyrics: line and put your words under [Verse] and [Chorus] tags. In our tests Lyria 3.5 sang every written line in order, and added a short ad-lib outro when time was left over.

How long can a Lyria 3.5 song be?

Up to about three minutes. Write the length or timestamped sections into the prompt; our three Lyria 3.5 tracks ended within two seconds of the requested length.

Can Lyria 3.5 make instrumental music?

Yes. Start with “Instrumental only, no vocals”. Our orchestral Lyria 3.5 cue returned no vocals and hit its planned timestamps within two seconds.

Which languages does Lyria 3.5 sing in?

English, Spanish, German, French, Portuguese, Hindi, Japanese and Korean. Write the lyrics in that language and name it in the style line.

What does a reference image do in Lyria 3.5?

It gives Lyria 3.5 a visual reference for mood and theme. Genre, length, structure and lyrics still come from the prompt.

How many credits does Lyria 3.5 use on Voor?

35 credits per track at any length, with or without an image. New accounts receive free welcome credits.

Can Lyria 3.5 copy a famous singer?

No. Lyria 3.5 blocks prompts that mimic real artists' voices. Describe the sound instead, and only submit lyrics you wrote or may use.

Should I use Lyria 3.5, MiniMax Music 3 or Mureka AI?

Lyria 3.5 for timestamp planning, multilingual vocals or an image-led score; MiniMax Music 3 for a duration number up to five minutes; Mureka AI for a set female or male lead or background music.

Plan your first Lyria 3.5 track

Write the sound in one paragraph, add a timestamp for each section, paste your lyrics under them, and compare two takes before you download.

Generate with Lyria 3.5