Google announced today that it is bringing AI music generation to the Gemini app, powered by DeepMind's Lyria 3 model. The feature is rolling out globally in beta and lets anyone generate an original 30-second song, complete with lyrics and AI-generated cover art, from a simple text description or an uploaded image.

What Is Lyria 3?

Lyria 3 is DeepMind's latest music generation model. Compared to its predecessors, it produces more realistic and compositionally complex tracks across a wide range of genres and moods. The model handles everything in a single pass: melody, arrangement, vocals, and lyrics are all generated together rather than stitched from separate systems.

The result is music that sounds cohesive rather than assembled. Google describes the output as significantly more expressive than earlier Lyria iterations, with better handling of tempo shifts, dynamics, and vocal performance.

How It Works

Users can prompt the tool in plain language. A description like "a melancholic jazz instrumental for a rainy evening" or "upbeat pop song about starting over" is enough to generate a track. The model produces the audio alongside matching lyrics and a cover image created using Google's Nano Banana image generation system.

The more interesting input mode is image and video-to-music. Upload a photo or short clip and Lyria 3 analyzes the visual mood, color palette, and subject matter to compose an appropriate soundtrack. This makes the tool immediately useful for content creators working with existing footage.

Availability

Music generation is rolling out to all Gemini users aged 18 and over. Supported languages at launch include English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese. The feature is live on the desktop version of Gemini now, with mobile availability expected over the next few days.

Free tier users are capped at 30-second generations. Google has not yet announced whether paid Gemini Advanced subscribers will receive longer or higher-quality outputs, though the current beta limits apply across the board.

YouTube Dream Track and SynthID Watermarking

Alongside the Gemini launch, Google is expanding Dream Track, its Lyria-powered music generation tool for YouTube creators, to a global audience. Dream Track was previously limited to select creators in the US.

Every track generated by Lyria 3, whether through Gemini or Dream Track, is automatically watermarked using Google's SynthID technology. SynthID embeds an imperceptible signal into the audio that survives compression, re-encoding, and most common post-processing operations. The watermark allows platforms and researchers to verify AI origin even after a track has been shared widely.

Why This Matters for AI-Generated Media

Music has been one of the harder modalities for AI to crack at consumer quality. Text and image generation reached mainstream adoption in 2023 and 2024 respectively. Video followed. Music has lagged, partly because of licensing complexity and partly because audio quality is harder to fake convincingly to a trained ear.

Lyria 3 landing inside Gemini, a product with hundreds of millions of users, changes the distribution equation entirely. This is not a research demo or a niche tool for producers. It is a mainstream feature available to anyone with a Google account. That scale accelerates the timeline for AI-generated music becoming a normal part of how people create content online.

What It Means for Video Creators

For creators working in video, the image-to-music feature is the most immediately practical capability. The ability to drop in a clip and receive a contextually appropriate soundtrack removes one of the most time-consuming parts of the post-production workflow: finding or licensing music that fits the mood.

This points to a broader shift in creative production. Every element of a video, from character performances to the soundtrack, is increasingly within reach of AI-powered tools. The barriers between an initial idea and a finished, polished video are collapsing faster than most people expected.