Google launches Lyria 3.5: Everything you need to know about its AI music model

In the latest update, Google has reportedly made Lyria 3.5 available in the Gemini app and API-a significant move that enables users to generate full songs at 44.1kHz from texts and image prompts with SynthID watermarking.

Google announced that it has integrated its Lyria 3.5 music model into the Gemini app and Gemini API. The model is capable of generating full songs with verses, choruses and bridges, launched in Google Flow on July 29.

However, the API produces 44.1kHZ stereo from significant texts and images. According to developers, users can upload up to ten images, structure lyrics using section tags, and apply timestamps to signal when specific instruments are introduced.

While the Lyria 3 Clip retains the ability to produce 30-second clips, Lyria 3.5 specifically used tracks shortened to roughly two to three minutes for broadcast.

The algorithmic memorization contained in model versions stored on German servers prevents software from relying on the statutory text and data mining exception for copying digital files.

The Lyria market is already overcrowded with more than half of new uploads to Deezer are AI-generated.

According to reports, around 90,000 fully AI tracks daily, accounted for between 1% and 3% of actual listening. Nonetheless, every Lyria track has its audio embedded with a SynthID digital signature that Google calls undetectable; moreover, Suno reached the same conclusion and scrambled to watermark the flood.