Lyria 3: Google's AI Music Generation Models Now Available via Gemini API
Gemini CLIGoogle launched Lyria 3, its most capable AI music generation system, via the Gemini API on March 25, 2026. The launch introduces two models — lyria-3-clip-preview for 30-second clips ($0.04 per clip) and lyria-3-pro-preview for full-length songs up to three minutes ($0.08 per song) — both accessible through the standard generateContent endpoint using any Gemini API key. The models accept text and image inputs, generate 48kHz stereo audio with structural awareness (verses, choruses, bridges), and watermark all output with Google's SynthID system for AI content identification. Lyria 3 was trained exclusively on licensed and permissible data, a deliberate differentiator from competitors Suno and Udio that have faced copyright litigation.
Sources & Mentions
5 external resources covering this update
Lyria 3 Comes to the Gemini API
Google introduced Lyria 3 — its latest generation of AI music models — via the Gemini API on March 25, 2026, making professional-grade music generation accessible to any developer holding a Gemini API key. The launch marks the most significant expansion of the Gemini API's creative capabilities to date, adding high-fidelity audio generation to a platform previously focused on text, code, and image understanding.
Two Models, Two Use Cases
The launch includes two model variants designed for different production contexts:
- lyria-3-clip-preview generates 30-second music clips at $0.04 per generation. It is optimized for speed and high-volume requests — suitable for background music in apps, short video accompaniments, and rapid prototyping.
- lyria-3-pro-preview generates full-length songs up to approximately three minutes at $0.08 per song. This model has deeper structural awareness and can produce compositions with professional song architecture including intros, verses, choruses, bridges, and outros.
Both models produce 48kHz stereo audio, with output available in MP3 and WAV formats.
How It Works
Lyria 3 integrates into the standard Gemini API using the familiar generateContent method, requiring no new SDK or special endpoint. Developers pass their API key as normal and specify either lyria-3-clip-preview or lyria-3-pro-preview as the model ID.
Both models accept multimodal inputs: text prompts describing the desired music, and up to 10 images that influence the mood, style, and atmosphere of the generated audio. Developers can exercise fine-grained control over the output by including structural tags directly in their prompts — [Verse], [Chorus], [Bridge] — along with specifications for tempo (BPM), musical key, genre, instrumentation, and lyrics with timing metadata.
Responsible AI and Licensing
Google emphasized that Lyria 3 was trained exclusively on licensed and permissible data. Every generated track is automatically marked with a SynthID watermark — imperceptible to human hearing but detectable by Google's verification tools — to identify AI-generated audio across distribution channels.
Availability and Pricing
Lyria 3 models are available in paid preview only — the free tier does not include access. Pricing is straightforward:
- Lyria 3 Clip Preview: $0.04 per generated clip
- Lyria 3 Pro Preview: $0.08 per generated song
Testing is available through Google AI Studio, and enterprise access is also available via Vertex AI.