Google on September 4, 2026, made Lyria 3.5, its music generation model, available in the Gemini app and the Gemini API, extending the model beyond the Google Flow Music tool where it debuted earlier this year. The model is available to all Gemini users globally on the web and in the mobile app, the company said in its announcement on The Keyword.
Google describes Lyria 3.5 as its best-sounding music generation model and said it brings more expressive vocals and richer musical arrangements than earlier versions, producing tracks at higher fidelity. The model first launched in Google Flow Music on July 29, 2026, when Google said it delivered improvements across musicality, lyrics, and vocal quality.
New Creation Controls in the Gemini App
The Gemini app version of Lyria 3.5 adds a set of guided creation features. Users can select or describe a genre and choose between vocal or instrumental styles. New templates are designed to start projects ranging from background music to custom birthday tracks, and users can now choose between short or longer tracks.
Google said the feature is aimed at use cases such as custom backing tracks for video, brand jingles, and personalized ringtones. Beyond the app, Lyria 3.5 is available for artists and AI creatives in Google Flow Music, and for developers and technologists through Google AI Studio and Google Vids.
Google DeepMind’s model page for the Lyria family describes the model as supporting variable song lengths of up to three minutes, with the ability to generate cohesive tracks matching a requested duration, from a 60-second clip to a full three-minute song. The page says users can write their own lyrics or generate them around specific subjects or themes, and can define vocal styles and acoustic preferences. It also describes composing with images: users can upload an image and have the model transform it into a track.
Developer Access Through the Gemini API
According to Google’s developer documentation for music generation, updated September 4, 2026, Lyria 3.5 generates 44.1 kHz stereo audio from text prompts or images, with structural coherence that includes vocals, timed lyrics, and full instrumental arrangements. The documentation lists two models: Lyria 3 Clip, under the model ID lyria-3-clip-preview, which always produces 30-second clips for loops and previews, and Lyria 3.5, under the model ID lyria-3.5, which generates full-length songs with verses, choruses, and bridges lasting a couple of minutes.
Both models are accessed through Google’s Interactions API and output MP3 audio by default, with WAV output available for Lyria 3.5. The documentation states that developers can supply up to 10 images alongside a text prompt, and the model will compose music inspired by the visual content. They can also provide custom lyrics using section tags such as Verse, Chorus, and Bridge, and use timestamps to specify what happens at particular moments in a song, such as when instruments enter. Prompting in a given language produces lyrics in that language, the documentation says, with the model adapting vocal style and pronunciation to match. Before audio is generated, the model reasons through musical structure based on the prompt.
Stated Limitations and Watermarking
The developer documentation lists several limitations. All prompts pass through safety filters, and requests for specific artist voices or the generation of copyrighted lyrics are blocked. Music generation is a single-turn process, so iterative editing of a generated clip across multiple prompts is not supported in the current version. Results may vary between calls even with the same prompt.
All generated audio carries a SynthID watermark for identification, which Google describes as imperceptible to the human ear. DeepMind’s model page states that all tracks are imperceptibly watermarked with SynthID so listeners and platforms can detect whether music was created or edited using AI, and that Google uses extensive filtering and data labeling to reduce harmful content in training datasets and the likelihood of harmful lyrics. The page adds that Google is still working on improving key capabilities and advises users to check that generated tracks fit their vision.
DeepMind’s page also notes that the Lyria line was developed with input from producers and musicians, and it lists earlier artist experiments with Google’s music tools, including producer Wyclef Jean’s work with Music AI Sandbox and Yung Spielburg’s use of Lyria in scoring the animated short Dear Upstairs Neighbors. Lyria 3.5 is available through the Gemini app, Google Flow Music, Google AI Studio, and Google Vids as of September 4, 2026.
Credit: Source link
