In brief
- Google’s Lyria 3 is rolling out in Gemini, generating full 30-second songs from text prompts or images.
- The model produces coherent tracks but struggles with niche genres and caps output length.
- Rivals like Suno and Udio still lead with longer songs, deeper controls, and more mature workflows.
Google has spent years quietly building its AI music model. On Tuesday, it finally put it somewhere everyone could actually use.
Lyria 3, Google DeepMind’s latest music generation model, is now rolling out in beta inside the Gemini app, letting any user over 18 describe an idea or upload a photo and get a fully produced track back in seconds—lyrics, instrumentals, and AI-generated cover art included.
“Just describe an idea or upload a photo, like ‘a comical R&B slow jam about a sock finding their match’ and in a matter of seconds, Gemini will translate it into a high-quality, catchy track,” Google said in its official blog. “To push the creative envelope further, you can even ask Gemini to take inspiration from something you upload.”
We gave it a spin. The short version: It works, it’s fun, and it may impress anyone who has never used other state of the art models like Suno or Udio. For those who have, it’s not going to replace their workflows anytime soon.
The tracks Lyria 3 produces are 30 seconds long. That’s the ceiling right now, and Google is upfront about it—the stated goal isn’t to make polished commercial songs but to create shareable moments. The output we tested was coherent, the lyrics matched the prompt, and the production quality was decent enough.
Here is what Google, Suno, and Udio can do with the same prompt:
The trouble starts when you push the edges. Lyria 3 seems to have a comfortable range of genres it navigates well—pop, afrobeat, R&B, light hip-hop. But when we tested prompts calling for more specific or unusual styles, the model had a hard time sticking to them. The prompting guide from DeepMind acknowledges this implicitly: It’s heavy on examples for mainstream genres and light on guidance for anything outside those lanes.
Compare that to Udio, which when it launched in 2024 came with controls like a “prompt strength” slider, a clarity adjuster for background noise, and negative prompting to exclude specific sounds or styles. Or Suno, which has been generating full multi-minute songs with proper song structure—verses, choruses, bridges—and is considered the best model in the scene. Both tools let you generate long-form tracks that feel like actual songs, not samples. Lyria 3’s 30-second cap and tendency to drift from unusual prompts put it in a different tier.
There’s also some legal context worth mentioning. Both Suno and Udio were sued by the Recording Industry Association of America in 2024 for allegedly training their models on copyrighted recordings without permission. Udio settled with Warner Music in November 2025 and is now transitioning to a fully licensed platform, launching sometime in 2026. Suno’s case continues.
Google, for its part, says it has been “very mindful of copyright and partner agreements” in training Lyria 3, and the model deliberately avoids mimicking specific artists—if you name one in a prompt, Lyria takes it as mood inspiration rather than a direct instruction.
On the safety and transparency side, all tracks generated in Gemini come embedded with SynthID, Google’s imperceptible AI watermark. The company also added audio verification to Gemini: you can upload a track and ask whether it was made by Google AI. That kind of provenance tooling is increasingly important as AI-generated audio floods streaming platforms—Deezer has already deployed detection tools to identify and flag fake streams from AI music.
Lyria 3 is available today on desktop for all Gemini users 18 and older in English, and a handful of additional languages. Mobile is rolling out over the next few days. Google AI Plus, Pro, and Ultra subscribers get higher generation limits. The model is also expanding YouTube’s Dream Track feature globally, previously U.S.-only, giving Shorts creators access to AI-generated soundtracks for their videos.
Daily Debrief Newsletter
Start every day with the top news stories right now, plus original features, a podcast, videos and more.
