The Pulse
Google Brings Lyria 3.5 Music Generation to the Gemini API
Google has released Lyria 3.5 in the Gemini API, adding full-length song generation with vocals, lyrics and structured arrangements. The model is also available in the Gemini app, Google AI Studio, Flow Music and Google Vids.
AI.info Team ·
Lyria 3.5 moves from Google’s music tools into the Gemini API
Google is making Lyria 3.5 available through the Gemini API, giving developers access to a music-generation model designed for complete songs rather than short audio fragments. The release, announced on September 4, also brings the model to the Gemini app, Google AI Studio, Flow Music and Google Vids.
Google describes Lyria 3.5 as its best-sounding music model, with more expressive vocals and richer arrangements. The company says the model can produce tracks with higher fidelity while preserving musical structure across verses, choruses and bridges.
“We have been developing our music generation tools in close partnership with industry experts to ensure AI serves as an additive force for human creativity,” wrote Alisa Fortin, product manager at Google DeepMind, and Guillaume Vernade, Gemini developer advocate at Google DeepMind, in Google’s March announcement of the Lyria 3 developer platform.
The Gemini API gets full songs, not just 30-second clips
The Gemini API documentation identifies lyria-3.5 as a preview model for full-length songs lasting a couple of minutes. Developers can influence the duration through a prompt, request a defined sequence such as an intro, verse, chorus and bridge, and specify details including genre, tempo, key, instruments and vocal style.
Lyria 3.5 accepts text or image input and returns MP3 audio by default, with WAV available through the API’s response-format setting. Google’s documentation says the output can include generated lyrics and a description of the song structure alongside the audio, allowing an application to handle the musical result and its textual components together.
Google is offering the model through its Interactions API. A developer can send a prompt such as a request for a cinematic orchestral track or a two-minute pop song, then read the returned audio and text blocks from the interaction response.
Google keeps Lyria 3 Clip for fast previews
The API now separates short-form experimentation from full-song generation. The Lyria 3 Clip model, identified as lyria-3-clip-preview, produces fixed 30-second MP3 clips for previews, loops and other short assets.
Both models support text and image inputs and produce 44.1 kHz stereo audio. Google recommends starting with the Clip model when testing prompts, then moving to Lyria 3.5 when a project needs a longer composition with distinct sections.
The distinction matters for developers building products around music generation. A short clip can support a social post, prototype or background loop, while the longer model can supply a complete arrangement with lyrics, vocals and instrumental changes that unfold over time.
Gemini adds templates and length controls for non-developers
Google is also changing how Lyria works inside the Gemini app. Users can select or describe a genre, choose between vocal and instrumental styles, start from templates and choose shorter or longer tracks. Google positions the feature for background music, brand jingles, birthday songs and ringtones.
Lyria 3.5 is available globally on the web and in Gemini’s mobile app, according to Google. The same model is available to developers through Google AI Studio and to users of Google Vids, while artists and other creative users can access it in Flow Music.
The company first introduced Lyria 3.5 in Flow Music on July 29, describing improvements to melody, lyrics, vocals and control over tempo and duration. Google said the updated model produces more natural melodic structures, more expressive vocal performances and improved pronunciation.
Prompt control becomes part of the product
Google’s developer documentation treats prompting as a form of arrangement. Users can specify section labels such as [Verse], [Chorus] and [Bridge], request a particular event at a specific time, or ask for lyrics in the language used in the prompt. Vocals and lyrics are generated by default, but prompts can request instrumental music or supply original lyrics.
The model does not currently support iterative editing within a single generation workflow, according to the documentation. Developers cannot generate a track and then refine it through multiple follow-up prompts in the same Lyria 3.5 process. Results can also vary between calls made with the same prompt.
Google’s September 4 release therefore expands access more than it changes the basic interaction model: developers describe a track, receive an audio file and accompanying text, and build their own editing or selection workflow around that output. The API’s model identifier is lyria-3.5, and Google lists it as a preview release.