Google Brings Lyria 3.5 AI Music Generation to Gemini

Google is rolling out Lyria 3.5, its latest AI music generation model, to the Gemini app, giving users a way to create complete songs, instrumentals and soundtracks from text prompts or uploaded images. The model, which had previously been available through Google’s Flow filmmaking tool, is also being made available through the Gemini API for developers.

Lyria 3.5 is designed to produce tracks with stronger vocals, more natural musical structure and better adherence to detailed prompts. Google says the model can generate songs as long as three minutes while maintaining a consistent flow from beginning to end.

Rather than producing only short repeating sections, the model can build out more complete arrangements with elements such as verses, choruses and bridges. Users can specify genres, tempo, dynamics, instrumentation and vocal characteristics, or begin with a simpler description such as an upbeat song for a particular occasion.

Gemini users can also choose between vocal and instrumental output and request either shorter clips or longer tracks. Templates are available for people who want a more guided starting point, including uses such as background music, jingles and personalized ringtones.

Lyria 3.5 also supports image-based composition. Users can upload a picture and ask the model to translate it into a custom piece of music, adding another input method alongside written prompts.

Google says the model can generate vocals in multiple languages and work across a wide range of genres. Users can provide their own lyrics or ask Lyria to create lyrics around a topic or theme, with the model then producing the corresponding vocal performance.

The company is also positioning Lyria 3.5 as a more controllable music-generation system. Prompts can include technical details such as beats per minute, key, instrument choices and vocal style, allowing users to shape the resulting track more precisely.

One limitation remains after generation: users cannot modify an existing track through the same prompt once it has been created. Google says it is still working to improve the model and recommends reviewing generated music to make sure it matches the intended result.

Lyria 3.5 is also arriving as AI-generated music becomes increasingly common on streaming platforms. Deezer has said AI-created tracks account for 44% of its daily uploads, while some streams associated with AI music have later been identified as fraudulent.

Google is using SynthID across Lyria output to distinguish AI-generated audio. Every track includes an imperceptible watermark that can be used to identify music created or modified with the company’s AI systems.

Google also says it uses filtering, data labeling and other safeguards intended to reduce harmful output. The system includes restrictions aimed at preventing users from cloning an artist’s voice or reproducing copyrighted lyrics.

With the Gemini rollout and API access, Lyria 3.5 moves beyond Flow into a broader set of Google products. That gives both consumers and developers direct access to a model capable of generating longer, structured music from text and images while giving users more control over vocals, style and composition.

See more here:

This analysis is based on reporting from Google DeepMind.

Image courtesy of Google.

This article was generated with AI assistance and reviewed for accuracy and quality.