Model comparison
Lyria 2 vs Lyria 3: Google's Music Models Compared
Lyria 2 (model id lyria-002) is Google's earlier text-to-music model on Vertex AI — instrumental only, roughly 30-second clips. Lyria 3 moved the family into the Gemini app in February 2026, and Lyria 3 Pro extended it to structured songs of up to about three minutes. Here is what changed and who still uses Lyria 2.
All values below come from Google's documentation and announcements; gaps are marked as not documented.
Background
What is Lyria 2?
Lyria 2 is an earlier Google music model exposed on Google Cloud Vertex AI under the model id lyria-002. Per Google's documentation it is text-to-music and instrumental only — no vocals and no lyrics. It produces clips of roughly 30 seconds (the docs list a maximum of 32.8 seconds) in WAV format at 48 kHz, accepts prompts in US English, and supports negative prompts to steer away from unwanted elements, with up to four clips per prompt.
Compared with the Lyria 3 generation, Lyria 2 has no image input and no song-structure controls. It remains a practical option for teams that already call it through Vertex AI and only need short instrumental beds.
Head-to-head
Lyria 2 vs Lyria 3 vs Lyria 3 Pro at a glance
The table only uses values Google has published. Where Google has not documented a capability, the cell says so instead of guessing.
| Dimension | Lyria 2 (lyria-002) | Lyria 3 | Lyria 3 Pro |
|---|---|---|---|
| Vocals and lyrics | Instrumental only — no vocals or lyrics | Can follow lyrics provided in the prompt, per Google | Can follow lyrics; built for structured songs |
| Track length per Google | ~30-second clips (docs: max 32.8 s) | About 30 seconds | Up to about 3 minutes |
| Song-structure prompting | None | Not documented by Google | Yes — intros, verses, choruses and bridges |
| Image input | Not supported | Not documented by Google | Supported — Gemini API docs list image inputs |
| Output format | WAV, 48 kHz | Not documented by Google | 48 kHz stereo, per Gemini API docs |
| Negative prompts | Supported | Not documented by Google | Not documented by Google |
| Where Google offers it | Google Cloud Vertex AI | Gemini app | Vertex AI (public preview), Google AI Studio, Gemini API, Gemini app (paid subscribers), Google Vids |
Still relevant
Who still uses Lyria 2?
Lyria 2 has not disappeared — it still serves pipelines that were built on Vertex AI before the Lyria 3 generation existed.
Vertex AI integrations
Teams that already call lyria-002 from backend jobs keep it because the integration, billing and prompt templates are stable.
Instrumental-bed workflows
When the brief only needs a short instrumental bed, Lyria 2's ~30-second instrumental output is sufficient and predictable.
Batch clip generation
With up to four clips per prompt and negative-prompt control, Lyria 2 suits workflows that generate variants at volume.
Migration
Moving a Lyria 2 prompt to Lyria 3
If you have Lyria 2 prompts worth keeping, a few habits translate cleanly to the Lyria 3 generation.
Describe what you want
Lyria 2 leans on negative prompts to exclude elements. With Lyria 3, prefer describing the target sound directly instead of listing exclusions.
Keep it instrumental-friendly
Instrumental prompts still work well in Lyria 3; in our studio you can also choose a vocal direction when the brief calls for it.
Use structure when you upgrade
If a clip should feel like a real arrangement, Lyria 3 Pro accepts structure words like intro, verse, chorus and bridge — something Lyria 2 never offered.
This site does not offer Lyria 2 generation. Our studio exposes the Lyria 3 family — Lyria 3, Lyria 3 Pro and Lyria 3 Clip — so the buttons on this page take you to Lyria 3. For Lyria 2 itself, use Google Cloud Vertex AI.
FAQ
Lyria 2 FAQ
What is Lyria 2?
Lyria 2 is an earlier Google music model available on Google Cloud Vertex AI as lyria-002. It is text-to-music, instrumental only, and generates clips of roughly 30 seconds in WAV at 48 kHz.
Can Lyria 2 generate vocals or lyrics?
No. Google documents Lyria 2 as instrumental only. Google describes the Lyria 3 generation as able to follow lyrics you write into the prompt.
How long are Lyria 2 clips?
Around 30 seconds; Google's docs list a maximum of 32.8 seconds per clip, with up to four clips returned per prompt.
Can I try Lyria 2 on this site?
No. Lyria3 AI offers the Lyria 3 family — Lyria 3, Lyria 3 Pro and Lyria 3 Clip — not Lyria 2. For Lyria 2 itself, use Google Cloud Vertex AI.
What did Lyria 3 Pro add over Lyria 2?
Per Google: tracks up to about three minutes instead of ~30-second clips, song-structure prompting (intros, verses, choruses, bridges), image inputs per the Gemini API docs, and availability across the Gemini app, AI Studio, the Gemini API, Vertex AI and Google Vids.
Is Lyria 2 free?
Lyria 2 runs on Google Cloud Vertex AI, where access and billing are decided by Google Cloud — we do not state Google's pricing. On this site, Lyria 3 generation uses our own credit system.
Try the newer models
Generate with Lyria 3 instead
Lyria 3 is the current generation. Open the studio, describe a track, and generate in the browser — Pro and Clip are in the same model menu.