Google DeepMind dropped Lyria 3.5 inside Flow Music on July 29, and the headline is the singing. This is a text-to-music model: type a prompt, feed it an image, or upload your own audio, and it generates a full song from 30 seconds to 3 minutes. The old knock on AI music was the vocals — robotic, mushy pronunciation. Lyria 3.5 goes straight at that, with more emotional, more natural voices and lyrics that actually follow the structure you ask for.
What you can actually do
You don’t just hit generate and pray. You talk to it: rework the song structure, add a bridge, tweak the mix, swap instruments — all conversationally. Tempo and length are directly controllable now, which was a real pain point before. Free tier gets you in; paid runs $8/$24/$64 a month.
Why it matters
Two things. It’s Google’s third Lyria model since February, so this is a product line, not an experiment. And unlike Suno and Udio — currently buried in Sony Music copyright suits — Google trains on licensed data. Lyria is available via the Flow Music workflow and the DeepMind Lyria model page for API access, so you can wire song generation into your own app.
You Might Also Like
- Google Lyria 3 Just Turned Gemini Into a Music Studio and im Weirdly Into it
- Google Deepmind Aletheia Just Solved Math Problems Nobody Could Heres why That Matters
- Suno ai 300m arr Milestone the ai Music app Thats Growing Faster Than Anyone Expected
- Elevenmusic Launches With Licensed Training Data While Suno and Udio Fight Copyright Suits
- 2 7b to Bring him Back Gone in 20 Months Noam Shazeer Leaves Google Deepmind for Openai
Top AI Product 
Leave a comment