Gemini creates AI music from text or photos

Gemini can now turn an idea, photo or video into a 30-second song. The feature is arriving in beta in Google’s app with Lyria 3, Google DeepMind’s new music generation model.

You only have to describe what you want to hear. It could be a song about a memory, an inside joke or your pet, with instructions about the genre, mood, voice and rhythm. You can also upload an image or video for Gemini to use as inspiration.

For example, you could ask for an afrobeat song about childhood memories with your mother, or a comedic R&B track about a sock searching for its partner. Within seconds, Gemini generates the music and, if you ask, writes the lyrics too.

More control and automatically generated lyrics

Lyria 3 improves three important areas compared with Google’s previous models:

• You no longer need to write your own lyrics: Gemini generates them from your prompt.
• You can define the style, voices and tempo more precisely.
• The songs aim to sound more realistic and have a more complex musical structure.

The result is not meant to replace a full music production. Google presents it as a quick way to create a personalized soundtrack to share with friends or accompany a specific moment.

Each song includes cover art created automatically with Nano Banana. You can download the result or share it through a link.

YouTube Shorts get the feature too

Creators can try Lyria 3 through Dream Track, YouTube’s music tool. It is available in the United States and is starting to roll out to creators in other countries.

The update lets you generate lyrical verses or instrumental tracks for Shorts, with more control over the music’s mood and style. In practice, someone posting a cooking video, a travel clip or a comedic scene can create a track tailored to that content without searching for an external song.

How to identify an AI-generated song

All songs generated in Gemini include SynthID, an invisible watermark developed by Google to identify AI-created content.

Gemini can also analyze an audio, image or video file and check whether it contains that watermark. You can upload a file and ask whether it was generated with Google tools. The answer combines SynthID detection with the system’s own analysis, so it should not be treated as universal proof that any content was created with AI.

Limits on imitating real artists

Google says the feature is designed to create original expressions, not copy specific artists. If you mention a musician in the prompt, Gemini interprets it as a general reference to style or mood, not as an instruction to reproduce that artist’s voice or an existing song.

The company also uses filters to compare results with content that has already been published and allows users to report possible violations. Even so, it acknowledges that these measures are not infallible. The terms of use also prohibit infringing copyright, privacy and intellectual property rights.

Lyria 3 is available in beta to users over 18 in English, German, Spanish, French, Hindi, Japanese, Korean and Portuguese. The rollout starts on the desktop version and will reach the mobile app over the following days. Google AI Plus, Pro and Ultra subscribers will have higher usage limits.

The important point is not that Gemini will replace a recording studio, but that creating a personalized song is becoming as simple as writing a sentence or uploading a photo. The next thing to watch is how Google expands the languages, creative controls and tools for clearly distinguishing music created by a person from music created by AI.