Google Launches Lyria 3: AI Music Generation Comes to Gemini

Google DeepMind has released Lyria 3, the company's most advanced AI music generation model to date. Announced on February 18, 2026, the model is now available inside the Gemini app, letting users create 30-second music tracks from text prompts or even images. It's a major step up from last year's Lyria 2 — and the first time Google has brought full music generation directly into its flagship AI assistant.

Video: Google DeepMind

What Lyria 3 Can Do

At its core, Lyria 3 turns a short text description into a complete music track. You type something like "upbeat jazz with a warm saxophone lead" and get back a 30-second clip with instruments, arrangement, and production that actually sounds coherent. The model handles multiple instruments, layered arrangements, and genre mixing without falling apart.

The biggest upgrade over Lyria 2 is vocal generation. Lyria 3 can now produce tracks with singing and lyrics automatically. Previous versions required users to supply their own lyrics — that's no longer the case. The model writes and performs them based on the mood and style you describe. You can still provide custom lyrics if you want more control, but the default automatic generation is surprisingly capable.

Image-to-music is another new trick. Upload a photo — a sunset, a cityscape, a party — and Lyria 3 will interpret the visual mood and generate a matching soundtrack. It's a niche feature, but it works well for content creators who want quick background music that matches their visuals.

Quality and Complexity

Google says Lyria 3 produces more complex and higher-quality output than its predecessor. In practice, this means better instrument separation, more natural dynamics, and arrangements that don't sound like they were generated by a machine stuck on loop. The tracks have real structure — intros, builds, and outros that feel intentional rather than random.

The model also handles genre requests more accurately. Ask for lo-fi hip hop and you'll get something that actually sounds like lo-fi hip hop, not a vaguely chill electronic track. This genre fidelity was a common complaint with Lyria 2, and the improvement is noticeable.

Availability and Languages

Lyria 3 launched first in the United States and is now rolling out globally. Music generation in the Gemini app is available to all users aged 18 and older, with support for prompts in English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese.

YouTube creators are getting a separate integration. Lyria 3 powers a new soundtrack generation feature for YouTube Shorts, letting creators generate custom background music directly within the Shorts creation flow. This is rolling out to creators in multiple countries.

SynthID Watermarking

Every track generated by Lyria 3 is embedded with SynthID, Google DeepMind's imperceptible audio watermark. This means AI-generated music can be identified programmatically, even after compression or minor edits. It's Google's answer to the growing concern about AI-generated content being passed off as human-made — and it applies to every single output from the model, with no way to disable it.

What This Means for Music AI

Lyria 3 is Google's clearest statement yet that AI music generation is ready for mainstream use. By embedding it directly in the Gemini app — the same app millions of people already use for chat, search, and writing — Google is betting that casual music creation will become as common as asking an AI to draft an email.

The quality bar has clearly risen. Lyria 3 doesn't replace professional music production, but for quick content soundtracks, social media videos, and personal projects, it's more than good enough. And with the YouTube Shorts integration, Google is making sure creators have a reason to actually use it.

Lyria 3 is available now in the Gemini app.