google / Music
Lyria 3 Pro
google/lyria-3-pro-previewFull music tracks with vocals or instrumentals, up to 184 seconds; actual length depends on the prompt and generated result.
Best for: Generating music tracks from text or images
Catalog status
Enabled
Research checked
2026-10-06
Benchmark results
Results are source-attributed, not TokenRex measurements. Different tests and thinking variants are not interchangeable.
1 of 1 results
Artificial Analysis AA-Music-Instrumental v1.1 Leaderboard
Media · Third-party report
Test conditions and source
Instrumental-only music generation preference Elo; distinct from Lyria 2 (Elo 932) and Lyria 3.5 (Elo 1076), which are different catalog entries not to be conflated with Lyria 3 Pro. Published source text: Leaderboard lists 'Lyria 3 Pro | Elo 1078±14 (95% interval 1064-1092) | 2,227 samples', ranked 3-6 behind Suno v6, Suno v6-mini, Mureka V9/V9.5.
Tested model: google/lyria-3-pro-preview
Artificial Analysis — AA-Music-Instrumental v1.1 Leaderboard · Checked 2026-10-06
Model specifications
Input token limit
131.07K
Published provider limit
Track duration
Up to 184 seconds (3:04)
Documented capability
Audio response
MP3
Documented capability
Input and output modalities
Provider-reported formats; API compatibility is detailed below.
Input
Output
- Exact model ID
- google/lyria-3-pro-preview
- Model type
- Music
- Input token limit
- 131,072 tokens
- Track duration
- Up to 184 seconds (3:04)
- Audio response
- MP3
- Prompt limit through TokenRex
- 4,000 characters
- Length control
- Describe the length in the prompt; no duration parameter
Input token limits describe provider input capacity. Published provider limits are not independently tested TokenRex request limits.
Versions and thinking levels
One enabled version is currently listed for this model.
Lyria 3 — everything it can do
Lyria 3 is Google's music-generation family. Clip is optimised for 30-second clips, loops and previews; Pro composes complete songs up to about 3 minutes with intros, verses, choruses and bridges. Both take text or an image as inspiration, sing vocals from generated or supplied lyrics, and return 48 kHz stereo MP3. On TokenRex you pay the official price, per track.
Clip vs Pro
| Feature | Lyria 3 Clip | Lyria 3 Pro |
|---|---|---|
| Model ID | google/lyria-3-clip-preview | google/lyria-3-pro-preview |
| Main use | Short clips, loops, previews | Full songs |
| Duration | 30 seconds (fixed) | Up to ~3 minutes |
| Song structure | Limited / short | Strong — intro, verse, chorus, bridge, outro |
| Market price | $0.04 per clip | $0.08 per song |
| TokenRex price | $0.04 per clip | $0.08 per song |
Capabilities
| Text → music | Yes |
| Image → music (set the mood from a picture) | Yes |
| Vocals | Yes |
| Lyrics generation | Yes |
| Custom lyrics with section tags ([Verse], [Chorus], [Bridge]) | Yes |
| Instrumental-only tracks | Yes |
| BPM control | Yes |
| Key / scale control | Yes |
| Genre, mood and energy control | Yes |
| Instrument selection | Yes |
| Arrangement and timing of musical events | Yes |
| 48 kHz stereo MP3 output | Yes |
| Batch API | No |
| Streaming | No |
| Function calling | No |
| Structured output | No |
| Live API | No |
How to control the music
Everything is controlled through the prompt text — there are no separate parameter fields. Google recommends specifying genre, instruments, BPM, key/scale, mood and arrangement for controllable results. You can also describe the vocal style, ask for instrumental-only output, and time musical events.
heavy hi-hats, cinematic strings, huge chorus, instrumental intro, drop at 20 seconds
For custom lyrics, paste them with section tags — [Verse 1], [Chorus], [Verse 2], [Bridge], [Outro] — and Lyria sings them in place. For image-to-music, send a picture (a cyberpunk city becomes a cyberpunk soundtrack; a mountain landscape becomes orchestral).
Good to know
- SynthID watermark: every generated track carries an inaudible SynthID audio watermark identifying it as AI-generated.
- Safety filtering: requests involving specific artist voices or copyrighted lyrics are filtered by the provider.
- One call, one track: each request returns one MP3. A failed request is refunded; the exact charge is in the
x-tokenbaazar-cost-paiseresponse header. - Quality: Google's model card reports improved audio fidelity and prompt adherence over Lyria 2 across music quality, aesthetics, vocal quality and fidelity. No independent public leaderboard score exists, so none is shown.
Model IDs:
google/lyria-3-clip-previewgoogle/lyria-3-pro-previewPOST /v1/music — see the Docs music section for copy-ready code.Features and tool support
TokenRex’s exposed interface, not every feature advertised by the provider. “Available” describes the implemented interface, not a successful test of every input or tool.
Input through TokenRex
AvailableText prompt, optional lyrics with [Verse]/[Chorus] tags and an optional image on POST /v1/music.
Output through TokenRex
AvailableOne 48 kHz stereo MP3 track per call.
Function tools
Not exposedMusic endpoint only — it creates songs and does not run tools.
Track duration
LimitedUp to 184 seconds (3:04)
Audio response
LimitedMP3
Prompt limit through TokenRex
Limited4,000 characters
Length control
LimitedDescribe the length in the prompt; no duration parameter
Sources and verification
Provider specifications describe the upstream model, not a guarantee of every feature through TokenRex. Prices come from the enabled catalog; benchmark results belong to the exact tested model.
- Vendor model documentation
Vendor-reported
Last checked - Lovable AI: music generation
Vendor-reported
Last checked - Google Cloud · Lyria 3 duration and audio specifications
Vendor-reported
Last checked - Artificial Analysis — AA-Music-Instrumental v1.1 Leaderboard
Third-party report
Last checked
Research notes (3)
- The provider publishes a 131,072-token input limit, but TokenRex accepts music prompts of at most 4,000 characters. These are different limits, not conflicting measurements.
- The provider supports lyrics text, but TokenRex returns the audio response only; a separate lyrics field is not exposed.
- Duration is documented by Lovable and Google, not a promise that every generated file is exactly that length. This audit did not generate or measure a new track.