Google Launches Lyria 3.5 AI Music Model Directly Inside the Gemini App

Google has released Lyria 3.5, its newest AI music generation model, and built it directly into the Gemini app. The rollout also extends to Google Flow Music, Google AI Studio, and Google Vids, giving both consumers and developers access to the model.

Inside the Gemini app, users can select a genre and style, choose between vocal or instrumental tracks, and generate short or extended compositions. Google has added templates aimed at first-time users, including presets for background music and personalized birthday songs. In Flow Music, Lyria 3.5 comes with additional creative controls not present in the base Gemini app integration. Developers can access the model through Google AI Studio, and it is also embedded in Google Vids for video soundtrack generation.

According to Google, Lyria 3.5 produces more expressive vocals and richer musical arrangements compared to its predecessor. The company has not published benchmark scores, listener studies, or other quantitative measures to support this claim.

Google states that Lyria was trained exclusively on licensed content, positioning it as a contrast to Suno, whose music generation models have faced copyright lawsuits from major record labels over training data provenance. However, Google has not disclosed which datasets, labels, or catalogs it licensed, nor how many hours of audio were used in training. Without this transparency, the licensing claim cannot be independently verified.

No pricing details have been released for API access to Lyria 3.5 through AI Studio. Context window, parameter count, and training cutoff date have also not been disclosed by Google.

What this means

Google is following a familiar playbook: embed a generative capability directly into its highest-traffic consumer surface — the Gemini app — rather than launching it as a standalone product. That distribution advantage is significant. Suno and Udio built their user bases as dedicated destinations; Google can put music generation in front of hundreds of millions of existing Gemini users with zero switching cost.

The licensing claim is the more consequential detail, even though it's unverified. Suno and Udio are currently defending lawsuits from Universal Music Group, Sony Music, and Warner Music over allegations that they trained on copyrighted recordings without permission. If Google can substantiate that Lyria 3.5 was trained solely on licensed material, it gives the company a legal and marketing edge over competitors facing existential litigation risk. But "licensed content" is a vague claim without specifics — Google could mean anything from direct label partnerships to stock-music libraries with unclear rights chains. Until Google names sources or publishes an audit, this remains a company assertion, not a verified fact.

The absence of pricing and technical specs also matters. Enterprises evaluating Lyria 3.5 for commercial video or app soundtracks need clarity on usage rights, output licensing, and cost per generation before adopting it at scale. Google's push into music generation signals that audio is becoming a standard modality for foundation model providers, alongside text, image, and video — but the real test will be whether the licensing claims survive scrutiny once artists, labels, or courts start asking pointed questions.