Google brings its Lyria 3.5 music model to every Gemini user

Google has made Lyria 3.5 available in the Gemini app and API, generating full songs at 44.1kHz from text or images with SynthID watermarking. A Munich court ruled against Suno on 31 July over memorised works and infringing outputs, asserting jurisdiction over training done outside the EU.

Google has put its Lyria 3.5 music model into the Gemini app and the Gemini API. It is available to all Gemini users globally on web and mobile, Google said.

The model generates full songs with verses, choruses and bridges. It first launched in Google Flow Music on 29 July, and also runs in Google AI Studio and Google Vids.

The API produces 44.1kHz stereo from text or images, the documentation says. Developers can supply up to ten images, write their own lyrics using section tags, and use timestamps to specify when instruments enter.

Two models are listed. Lyria 3 Clip always produces 30-second clips, and Lyria 3.5 produces tracks lasting a couple of minutes. Generation is single-turn, so a clip cannot be edited across prompts.

The guardrails are specific. Safety filters block requests for particular artist voices and for the generation of copyrighted lyrics.

Those two prohibitions describe a European court case almost exactly. Google has not said the two are connected.

A Munich court found on 31 July that Suno had broken copyright, in the first ruling of its kind in Europe.

The works were memorised and reproducibly contained in model versions stored on German servers, the court held, which put them outside the text and data mining exception. It also held Suno responsible for the outputs, rejecting the argument that complex prompting shifted that responsibility to users.

It went further still. The court asserted jurisdiction over training done outside the EU, ran its own analysis of US fair use, and concluded it did not apply.

The injunctions cover training on the works, retaining them in model parameters and offering the trained models in Germany. The order carried language reaching copying inside the United States. The judgment is not final and can be appealed.

The market Lyria is entering is already saturated. More than half of new uploads to Deezer are AI-generated, and Deezer has been fighting back while Spotify has not.

Around 90,000 fully AI tracks arrive there daily, and they account for between 1% and 3% of actual listening. Up to 85% of the streams they generated in 2025 were fraudulent.

Every Lyria track carries a SynthID watermark that Google calls imperceptible. Suno reached for the same answer, moving to watermark the flood once the lawsuits began.