Lyria 3.5 Music Models Launch in Gemini API
Gemini CLIGoogle brought its next-generation Lyria 3.5 music generation models to the Gemini API in public preview, giving developers programmatic access to the same model that powers Google Flow Music. The release ships two variants: lyria-3.5-clip-preview for fast 30-second clips and loops, and lyria-3.5-pro-preview for full-length songs up to three minutes with improved musical coherence, more natural vocals, and finer structural control. Both models accept text and image prompts and generate 44.1 kHz stereo audio, replacing the Lyria 3 preview models as the new frontier tier for API-based music generation.
Featured Video
A video we selected to help illustrate this changelog
Key Takeaways
- Google shipped Lyria 3.5 to the Gemini API in public preview on September 3, 2026, giving developers direct API access to a model previously confined to Google Flow Music.
- The release splits into two purpose-built endpoints:
lyria-3.5-clip-previewfor fast 30-second clips andlyria-3.5-pro-previewfor full multi-minute songs with structural control. - Lyria 3.5 brings improved vocals and lyrics, generating more natural pronunciation and better prompt adherence than the prior Lyria 3 preview models.
- Both models support multimodal prompting, accepting up to 10 images alongside text to steer the generated audio's mood and style.
- Every output includes an inaudible SynthID watermark, and safety filters restrict prompts referencing specific artist voices or copyrighted material.
- Because Lyria trains on licensed data, it gives developers a comparatively lower legal-risk alternative to rivals like Suno and Udio, which face active copyright litigation.
Sources & Mentions
4 external resources covering this update
Google Flow Music is HERE! How to Use Lyria 3.5 (Free AI Music Generator)
YouTube
Google's Lyria 3.5 makes AI-generated songs sound more natural and expressive
Neowin
Google's Lyria 3.5 Sharpens Vocals and Lyrics While Rivals Fight in Court
Tech Times
Google DeepMind's Lyria 3.5 Takes Direct Aim at Suno's Vocal Dominance
AlphaSignal
Lyria 3.5 Comes to the Gemini API
Google has opened up Lyria 3.5, its newest music generation model family, to developers through the Gemini API in public preview. The models had previously been available only inside Google Flow Music, the company's consumer music creation surface; this release marks the first time Lyria 3.5's underlying capabilities are directly accessible via API for developers to build into their own products.
Two Models, Two Use Cases
The release introduces two distinct preview endpoints. lyria-3.5-clip-preview is optimized for fast iteration: it generates fixed 30-second clips and loops, aimed at rapid experimentation, previews, and short-form use cases. lyria-3.5-pro-preview is the flagship variant, built for full-length song generation with duration control that stretches to several minutes and explicit structural awareness of verses, choruses, and bridges.
Both models accept text prompts, and developers can additionally supply up to 10 images to influence the generated audio's mood and style. Output is high-fidelity 44.1 kHz stereo audio, delivered as MP3 by default, with Lyria 3.5 Pro also supporting WAV output for higher-fidelity workflows.
What Improved Over Lyria 3
Lyria 3.5 carries forward the musicality, lyrics, and vocal quality gains Google introduced when the model first launched inside Flow Music in July 2026. Compared to the earlier Lyria 3 preview models, Lyria 3.5 produces richer melodic structures, better prompt adherence for generated lyrics, and more expressive, naturally pronounced vocals. Developers can also specify custom lyrics with section tags, control tempo, BPM, and key, or generate instrumental-only tracks.
Access and Limitations
The models are accessed through the Interactions API, with sample code available in Python, JavaScript, Java, and REST. As with other Google audio and video generation models, every clip carries an inaudible SynthID watermark for provenance identification, and safety filters block prompts that request specific artist voices or copyrighted material. Music generation remains a single-turn process, meaning there is no iterative in-context editing of a generated track, and identical prompts can still produce varying results between runs.
Why It Matters
Lyria 3.5's arrival on the Gemini API gives developers a meaningfully upgraded alternative to Lyria 3 for programmatic music generation, at a moment when competition in AI music tools (Suno, Udio, and others) is intensifying. Because Lyria trains on licensed data rather than facing the copyright litigation some rivals are dealing with, it offers a comparatively lower-risk option for developers building commercial products around AI-generated audio.