A Beta Feature Turns Prompts Into 30-Second Songs
Google DeepMind’s Lyria 3 music-generation model has been folded into Gemini, letting people in the beta test produce 30-second songs starting from either written prompts or images they upload. Whatever the starting point — a sentence of text or a picture — the system converts it into a brief original song complete with custom cover artwork, adding a fresh creative function to what Gemini can do.
As Google’s strongest music-generation model yet, Lyria 3 is being introduced inside the Gemini app. According to the company, simply typing out a concept or attaching a photo is enough for Gemini to produce a track within moments. Google frames this less as a professional recording tool and more as a way for everyday users to express themselves through brief pieces of audio.
Prompts Can Come From Words, Pictures, or Clips
According to the material describing the tool, Gemini’s music generator can be driven not just by written prompts but also by photos and video clips. People can request a particular musical genre, a certain mood, a private joke between friends, or a personal memory, and depending on that input, Gemini will either produce a song with lyrics or an instrumental piece.
Google points to three upgrades that separate Lyria 3 from earlier versions of the model: automatic lyric writing, finer creative control over elements such as style, vocal delivery, and tempo, and the ability to produce audio that sounds more lifelike and musically intricate. Within Gemini, each finished song is also matched with cover art produced by Nano Banana, a pairing meant to make it simpler for people to share what they’ve made.
Google has offered sample prompts to illustrate the breadth of what the tool can handle — among them a humorous R&B slow jam built around a sock finally reuniting with its missing pair, a wistful afrobeat song touching on family memories and plantains, and a track inspired by nothing more than a snapshot of a dog out on a hike. Those examples underscore that the tool is aimed at lighthearted, personalized creations rather than polished studio production.
SynthID Watermarks and Age Rules Shape the Launch
Each song produced through Gemini carries SynthID, Google’s invisible watermarking technology used to flag content made by artificial intelligence. Google states that this embedded marker allows anyone to confirm afterward that a given piece of audio originated from AI.
Access is restricted to Gemini app users who are at least 18 years old, and the tool supports several languages, with paying subscribers granted higher usage limits. Per the source description, this is currently a beta launch, offered globally in eight languages: English, Japanese, Korean, Hindi, Spanish, Portuguese, German, and French.
Sharing a finished song is possible either by downloading the file directly or by sending out a share link, Google notes. Every track runs 30 seconds, and the company is upfront that the goal isn’t to craft a masterpiece — it’s to give people a fast, distinctive outlet for expressing an idea.
A Parallel Route Opens Up for YouTube Creators
Gemini isn’t the only place this technology is landing. Google notes that creators have another avenue for Lyria 3 through YouTube’s Dream Track feature, which aims to raise the quality of one-of-a-kind soundtracks for Shorts videos. Dream Track is currently live in the United States, and the source material indicates it’s expanding to YouTube creators in additional countries.
Lyria 3 also appears in Google’s developer-facing Gemini API documentation, which describes the model’s capacity to generate 44.1 kHz stereo audio of high quality from either text or image inputs. That same documentation notes that the models maintain structural coherence throughout a track, encompassing vocals, precisely timed lyrics, and complete instrumental arrangements.
Taken together, this rollout spreads Lyria 3 across both everyday consumer tools and creator-oriented products at Google, with Gemini serving as the most straightforward entry point for making short songs. The source material confirms this is a beta feature and that it has already gone live within the Gemini app.
For editorial consideration and industry coverage inquiries, contact [email protected]
