Google Adds Custom Music Generation to Gemini with New Lyria 3 Model

Google is expanding the creative ambitions of its Gemini app with the introduction of Lyria 3, a new generative music model that can turn text prompts, photos or videos into short, shareable tracks. The feature, rolling out now, aims to make music creation feel as casual as typing a message.

Users can describe almost anything, from a heartfelt memory to a surreal joke, and Gemini will produce a 30 second song complete with lyrics, vocals and cover art. Google says the system can also take cues from uploaded images or clips, using them as inspiration for tone and subject matter.

Lyria 3 is pitched as a clear improvement on the previous version. It now writes its own lyrics, offers more control over style, vocals and tempo, and produces audio that Google claims is more realistic and musically complex. The company frames the tool as a way to spark playful creativity rather than replace professional musicians.

Nano Banana and Lyria 3

The app supports two main modes. Text to tracks lets users request a specific genre or mood, such as an Afrobeats song about a favourite childhood meal. The second mode builds music from photos or videos, generating lyrics that match the scene, like a track about a dog exploring the woods.

Each piece comes with cover art generated by Nano Banana and can be downloaded or shared instantly. Google says the goal is to give people a quick, personal soundtrack for everyday moments.

Creators on YouTube Shorts will also see benefits. Lyria 3 is being added to Dream Track, first launched in the United States and now expanding to more regions. Google says the upgraded model will improve the quality of custom Shorts soundtracks, whether users want a short verse or a simple backing track.

Audio Watermark

To address concerns around AI generated audio, all tracks made in the Gemini app include SynthID. This is Google’s watermarking system for identifying AI content. Users can upload an audio file and ask Gemini to check whether it contains SynthID. The model then combines watermark detection and its own reasoning to provide an answer.

Google stresses that Lyria 3 has been developed with input from the music industry. They add that it is designed for original expression rather than imitation. If a user names a specific artist, the system will aim for stylistic influence rather than a direct copy. Filters are in place to check outputs against existing works, and users can report anything that appears to infringe rights. The company notes that its Terms of Service and generative AI policies prohibit violating intellectual property or privacy.

Lyria 3 is available now in the Gemini app for users aged 18 and over in eight languages,. These include English, German, Spanish and Japanese. Google has announced plans to expand both quality and language support over time.