Google launches Lyria 3 music generation model – News

Google LLC today introduced an artificial intelligence model called Lyria 3 that consumers can use to generate short tracks.

The algorithm is rolling out to the company’s Gemini app and Dream Track, a music generation feature in YouTube’s creator toolkit. Files generated with Lyria 3 will contain an imperceptible watermark generated by a Google technology called SynthID. Users can check whether a track contains the watermark by uploading it to the Gemini app.

Lyria 3 generates 30-second tracks based on natural language prompts. Users can specify details such as the genre of the track they wish to generate, its tempo and the language of the lyrics. Alternatively, they can upload an image or a video and have Lyria 3 automatically generate a matching tune.

The model offers several improvements over Google’s previous-generation Lyria 2 algorithm, which debuted last May. Users no longer have to bring their own lyrics because they’re created automatically. Additionally, Google has boosted the quality and complexity of the outputted music.

The way an AI model goes about generating music depends on its architecture. Some algorithms don’t produce audio directly from prompts, but first create an intermediate representation called a spectrogram. It’s a data visualization that represents tones as lines. Other models, such as Google’s open-source MusicML algorithm, represent music as compressed units of data called audio tokens.

The Gemini app enables users to create cover art for each track they generate. The feature is powered by Nano Banana, an AI image generator that Google debuted last year. It’s available not only through Gemini but also an application programming interface that developers can integrate into their software. It’s possible that Lyria 3 will likewise become accessible via an API down the road. 

In the meantime, the model is available to adult users of Gemini’s mobile client. Google plans to bring Lyria 3 to the desktop version in the coming days. Customers with Google AI Plus, Pro and Ultra subscriptions will have higher usage caps.

The launch could create more competition for AI music startups such as Suno Inc., which raised a $250 million round in November. The company provides a freemium service that makes it possible to generate audio with natural language prompts. Suno’s paid plans include additional features, including a virtual audio workstation that enables users to customize AI-generated tracks manually.

There are several ways Google could enhance Gemini’s music features in the future. It could extend the 30-second track length limit and roll out editing features similar to those provided by Suno. The company could also integrate Lyria 3 into more of its consumer services, such as by bringing AI-generated audio to the virtual worlds users generate with Project Genie.

Image: Google