Google has made Lyria 3.5 available in the Gemini app and the Gemini API, moving its latest music-generation model from a creative preview into a broader consumer and developer surface.
The model now matters for two different audiences. In Gemini, Google is pitching easier song creation with genre selection, vocal or instrumental styles, templates, and shorter or longer tracks. In the Gemini API, Lyria 3.5 becomes a programmable model for generating music from text prompts or image inputs.
The developer model page lists lyria-3.5 as the model code, with text and image inputs, MP3 audio output, lyric text output, and a 131,072 input-token limit. Google says the model generates high-quality 44.1 kHz stereo audio.
The API path is the bigger platform change
Google’s music-generation guide says Lyria 3.5 is available through the Gemini API using the Interactions API. The guide separates the family into two model paths: lyria-3-clip-preview for 30-second clips and lyria-3.5 for full-length songs that last a couple of minutes.
That split gives developers a practical product choice. Short clips are useful for loops, previews, and lightweight creative tools. Lyria 3.5 is positioned for songs with verses, choruses, bridges, vocals, timed lyrics, and full instrumental arrangements.
Google also documents multimodal input support. Developers can provide up to 10 images alongside text prompts, and the model can compose music inspired by the visual material. That moves the product beyond a text-only prompt box and closer to a creative workflow where visual direction, lyrics, structure, and audio generation sit together.
Capability claims still need evidence
Google’s Gemini app post says Lyria 3.5 brings more expressive vocals, richer arrangements, and higher fidelity. The DeepMind model card says Lyria 3.5 improved compared with Lyria 2 on audio fidelity and prompt adherence, especially when lyrics are involved.
Those are useful claims, but they remain Google-supplied. The practical test for creators and developers is whether Lyria 3.5 can produce controllable output that survives real editing, licensing review, brand constraints, and repeated prompting.
The model docs also define boundaries. The Lyria 3.5 model page lists audio generation as supported while marking several other Gemini capabilities as unsupported for this model, including code execution, file search, function calling, Search grounding, structured outputs, and Live API support.
The music model is becoming a Gemini surface
The notable shift is distribution. Lyria 3.5 is now available to all Gemini users globally on web and mobile, according to Google, and it is also available through Google AI Studio, Google Vids, Flow Music, and the Gemini API.
That breadth makes Lyria less like a standalone music demo and more like a capability Google can place across consumer creation, video workflows, and developer products. For The AI Feed readers, the question is not only whether the songs sound better. It is how quickly music generation becomes another programmable media primitive inside broader AI apps.
