Gemini no longer just generates images or videos. Now it can also turn an idea, a photo, or a video into a personalized 30-second song, complete with lyrics, vocals, rhythm, and cover art. The feature arrives thanks to Lyria 3, Google DeepMind’s new generative music model.
From an idea to a song in seconds
The concept is simple: describe what you want to hear, and Gemini creates a music track based on your prompt. Do you want an R&B song about a sock finding its match? An Afrobeat track to remember the meals from your childhood? The tool tries to turn that concept into a piece with its own lyrics and style.
You can also upload a photo or video. For example, a picture of your dog during a walk can become a song about that adventure. You don’t need to write the lyrics or know anything about music production: just explain the mood, genre, tempo, or emotion you’re looking for.
Lyria 3 is designed to help you express yourself quickly and have fun, not necessarily to replace a professional composer, producer, or singer.
According to Google, the model improves in three main areas:
- It generates lyrics from your description, without requiring you to provide them.
- It offers more control over the style, vocals, and tempo.
- It produces tracks with greater realism and musical complexity.
Each result includes cover art created with Nano Banana. You can then download the song or share it through a link, which is useful for sending a greeting, adding music to a family memory, or creating an inside joke among friends.
Music for Gemini and YouTube Shorts
Lyria 3 is also coming to Dream Track, a YouTube feature that lets you create soundtracks for Shorts. The rollout began in the United States and is expanding to creators in other countries.
The tool can generate anything from a lyrical verse to an instrumental track to accompany a video. For people who publish content frequently, this offers a more flexible way to adapt the music to the exact tone of each Short.
How to tell if a song was created with AI
Google says that every song generated in Gemini includes SynthID, an invisible watermark designed to identify content created with Google’s AI tools.
The app is also expanding its verification features. You can upload an audio file and ask Gemini whether it was generated with Google technology. The system will look for the SynthID watermark and combine that result with its own analysis to provide an answer.
This doesn’t mean detection is foolproof. The company itself acknowledges that its approach can make mistakes, so it’s best to treat these results as a signal rather than absolute proof of a file’s origin.
Limitations, copyright, and availability
Google presents Lyria 3 as a tool for creating original expressions, not for directly imitating existing artists. If you mention a musician in your prompt, Gemini should treat that as a general reference to a style or atmosphere, rather than copying their voice or producing an exact replica.
The company also says it applies filters to compare results with existing content and maintains agreements and measures related to copyright during the model’s training. Even so, it recommends reporting any content that may violate intellectual property or privacy rights.
Lyria 3 is beginning to roll out in the Gemini app for users over 18. It is available in English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese. The rollout starts with the desktop version and will reach the mobile app in the following days.
Google AI Plus, Pro, and Ultra subscribers will have higher usage limits. For now, each creation lasts 30 seconds, which is enough time to share an idea, accompany a short video, or add music to an everyday moment.
Music generation is starting to look less like an exclusive tool for professional studios. With Lyria 3, a random thought, a family photo, or a memory can become a small song in just a few seconds. The question is no longer whether you know how to compose, but what story you want to hear today.
Original source
https://blog.google/innovation-and-ai/products/gemini-app/lyria-3
