Music to your ears, literally: Gemini now writes and produces songs

Over the recent years, Gemini has really stepped up its game. The recent Gemini 3 update improved its photo and video quality to an unrealistically high standard, while the more recent deep thinking upgrade further enhanced its coding and research capabilities. After excelling at its text, image, video, and code generation, Gemini is taking major steps forward toward a relatively overlooked aspect: audio, or more precisely, music.

With today’s update, Gemini is taking a big leap in AI-powered music generation and enabling users to create music with a simple text or image prompt. This functionality is powered by Google DeepMind’s new and updated Lyria 3 text-to-audio model. While previous versions of Lyria have been limited to experimental tools such as MusicFX or PromptDJ, Google is now making it accessible to a much broader audience through the Gemini mobile apps and web interface, just as we spotted last month.

With this upgrade, Google says, users can describe a concept or upload an image, and Gemini will convert it into a “high-quality, catchy track.” The new feature will also generate lyrics while generating tracks that actually replicate real music, i.e., they feature a complex and melodious mix of various tastefully layered instruments. So far, apps such as Suno have dominated AI-generated music.

Here’s a preview of its capabilities:

Previous Lyria models seemed rather limited and produced samples that required human expertise for layering and organizing into a musical piece. It could also produce soundtracks through mediums, such as Dream Track for YouTube Shorts, or through the Recorder app on Pixel 9 and 10 series phones. However, those were also limited to instrumental clips intended to be fillers or background music.

With this update, Gemini can create 30-second tracks that will be accompanied by an album art created with Gemini Nano. Google adds that Gemini will also let users tweak different aspects of the track, including the tempo, or change the music or singing style per instructions. Users can also add existing music tracks for Gemini to take inspiration from.

Preventing unauthorized use of copyrighted tracks

Google has a firm stance against copying original music. Each track comes with a hidden SynthID to help platforms differentiate AI-generated music from that created by actual human artists. Google adds that it has been “mindful of copyright and partner agreements” while training the new model, and claims that it will not mimic existing artists. The output tracks will also be tested against existing musical content to avoid infringement. Users will also be able to report any seemingly copied music track that slips through cracks.

That becomes pertinent, especially as Google was recently alleged to have used a radio presenter’s voice for NotebookLM’s AI-generated podcast feature.

Unlike other Gemini experiments, Google is going all-in with music creation. It is launching the feature to Gemini in beta in several languages, such as English, French, Hindi, German, Japanese, Korean, Portuguese, and Spanish. In addition to Gemini, the feature will also be available through the Dream Track audio generator for YouTube Shorts.

The feature is available to both free and paying Gemini users, although available credits would depend on your subscription status.