Google brings Lyria 3.5 music generation to the Gemini API in public preview

Google has put Lyria 3.5, its music generation model, into public preview on the Gemini API, extending a model that debuted in Google's Flow Music tool in July to developers building it directly into their own products. The Gemini API changelog describes the addition as "full-length song generation with improved musical coherence, natural vocals, and fine-grained duration and structural control," and says the model accepts both text and image inputs.

What's new

Lyria 3.5 was first announced on July 29, 2026 inside Google's Flow Music app, where Google said it improved on prior Lyria versions with richer melodic structures, higher-quality lyrics with better prompt adherence, more emotionally expressive vocals with improved pronunciation, and finer creative control over tempo and track length. What changed this week is distribution: rather than being confined to Flow Music, the model is now reachable through the Gemini API, meaning developers can call it programmatically to generate music inside their own applications rather than only through Google's own consumer tool.

The API release keeps Lyria 3.5's core capabilities — generating structurally coherent full-length songs, including vocals and timed lyrics, from a single prompt, with tracks rangeable from around 30 seconds up to three minutes. Earlier Lyria versions and Google's other generative media models have shipped with SynthID audio watermarking baked into every output and guardrails against reproducing an identifiable artist's voice or copyrighted lyrics; Google has said the same protections carry over as its music models move from consumer surfaces to developer-facing APIs.

Context

Lyria has been Google DeepMind's answer to Suno, Udio, and ElevenLabs' Music in the fast-growing AI music generation space, and Google has moved quickly to widen its distribution since the 3.5 launch: from a single consumer app in July to the Gemini app and now the developer-facing Gemini API in September. That mirrors the path Google has taken with its other generative models — ship first in a dedicated consumer tool, then fold the capability into the general-purpose Gemini API once it is stable enough for third-party integration.

The move also lands as the music-generation category has drawn heightened scrutiny over copyright: rival labs and music publishers have sued competing AI music generators, and regulators and rights holders have pushed for clearer provenance and licensing controls on AI-made audio, which is part of why watermarking and artist-mimicry blocks have become standard features rather than optional add-ons across the major music models.

Why it matters

Opening Lyria 3.5 to third-party developers rather than keeping it locked inside Flow Music turns Google's music model from a standalone consumer product into infrastructure other companies can build on — game studios, video editors, and content platforms can now generate licensed-feeling soundtracks on demand through an API call instead of a dedicated app. It also puts Google in more direct competition with API-first music-generation vendors like ElevenLabs and Suno, who have built developer businesses around exactly this kind of programmatic access, at a moment when demand for AI-generated background music and scoring for video content is climbing alongside the broader boom in AI video generation.