Google launches Lyria 3 for AI music creation

Google is opening Lyria 3 to developers worldwide, allowing them to generate songs with vocals, lyrics and a complete musical structure from text or images. The model is available in public preview through the Gemini API and a new creation experience in Google AI Studio.

This is not just about producing a short melody. Lyria 3 is designed to maintain musical coherence throughout the entire piece, so the verses, choruses, vocals and style do not sound like disconnected parts.

Two models for two needs

Google offers two variants with different goals:

• lyria-3-pro-preview: generates songs of up to approximately three minutes. It is designed for projects that prioritize a more elaborate structure and studio-quality output.
• lyria-3-clip-preview: creates 30-second clips and is optimized for speed and a high volume of requests. It can be useful for prototypes, background loops or content for social media.

Both versions can create more realistic and expressive vocals, as well as music in different languages and genres. You can ask for anything from pop and funk to Motown-inspired styles, as long as the request falls within what the model understands.

More control over the song

Lyria 3 also includes controls for directing generation with natural-language instructions. You do not have to limit yourself to writing “create a happy song” and waiting for the result.

For example, you can specify:

• The tempo, so the song is fast or slow and fits the rhythm of an app or video.
• When the lyrics begin and end within the track.
• The song’s progression, including an introduction, verses, choruses and bridges.
• An image to define the atmosphere, style or emotion of the music.

Image input lets you turn a visual reference into a guide for the sound. A nighttime photograph, for example, can steer the creation toward a darker or more atmospheric mood.

A space to try it in AI Studio

Google is also launching a dedicated environment in AI Studio. With a paid API key, developers can experiment with Lyria 3 without first building a complete application.

The space includes two creation modes:

• Text mode: describe the music and add parameters such as tempo or key.
• Composer mode: build the song section by section and adjust each part’s duration, intensity and description.

This could speed up the development of tools for creating soundtracks, music for videos, social media content or game prototypes. For the general public, the most important change is that these features may start appearing inside third-party applications, not just Google products.

Markers to identify generated music

Every track created with Lyria 3 includes a SynthID digital watermark. According to Google, this technology makes it possible to identify and verify that audio was generated by its AI even after it has been modified.

Lyria 3 Clip and Lyria 3 Pro are now globally available to developers in public preview. The next thing to watch is how they perform in real products: generating a convincing song is only part of the challenge. Maintaining control over vocals, rights and the use of the music will be just as important.