Google’s official account announced on X on September 4 that its most powerful music generation model, Lyria 3.5, is now officially available on the Gemini App, allowing all users worldwide to generate full songs directly on mobile or web. The model first appeared in Google Flow Music at the end of July, and it took a month before it was rolled out to Gemini, a larger-scale entry point. For regular users, the difference is that they no longer need to switch to Flow Music—they can generate music directly within the Gemini conversation.
From Flow Music to Gemini: The Complete Path of Lyria 3.5
Lyria 3.5 first launched on July 29 on Google Flow Music, where Google highlighted its improvements in musicality, lyric quality, and vocal performance. On September 4, this update brought it into the Gemini app and Gemini API, expanding coverage to web, mobile, Google AI Studio, and Google Vids. This also marked the first time Lyria 3.5 left Flow Music, a relatively niche entry point.
In terms of the model’s capabilities: Lyria 3.5 can generate 44.1kHz stereo audio, producing complete song structures that include verses, choruses, and bridges, rather than just repeating a short melody loop. Lyria 3.5 shows improvements in melodic structural complexity, prompt adherence, structural awareness, and pronunciation accuracy, with vocals that sound more realistic and emotionally nuanced. The DeepMind model page shows that Lyria supports song generation of up to three minutes, and users can customize lyric themes, vocal styles, and audio preferences, as well as upload images for the model to generate music based on them.
New ways to interact in the Gemini App
This launch of the Gemini App brings three new feature sets. First, templates: Google provides a series of starting templates covering scenarios such as background music, brand jingles, and personalized ringtones, so users don’t have to start from a blank prompt. Second, track length selection: users can choose short clips or longer full tracks. Third, style control: users can directly select or describe the music genre, and switch between tracks with vocals and purely instrumental versions.
Google’s official blog clearly lists the target use cases: background music for videos, brand jingles, and personalized ringtones. Lyria 3.5 works within Gemini the same way as the previous Flow Music, but it adds these guided tools, which are especially useful for beginners who aren’t familiar with writing prompts.
The specifications on the API side are more detailed. The Gemini API documentation lists two model IDs: `lyria-3-clip-preview` always outputs 30-second clips, while `lyria-3.5` generates complete tracks lasting several minutes. Developers can provide up to ten images along with the prompt, use section tags (Verse, Chorus, Bridge) to write custom lyrics, and use timestamps to specify when instruments enter. The default output is MP3, with Lyria 3.5 additionally supporting WAV format. The documentation also explains that whichever language is used in the prompt, the generated lyrics will be in that language, and the vocal style and pronunciation will adjust accordingly.
Lyria 3.5, our best-sounding music generation model, is now available in the @GeminiApp 🎵✨
With more expressive vocals and richer musical arrangements, it’s easier than ever to bring your idea to life:
🪄 Use our new templates to jumpstart your creativity
⌛ Choose to create… pic.twitter.com/ZcwcsWH91J— Google (@Google) September 4, 2026
SynthID and Copyright Guardrails
All audio generated by Lyria comes with SynthID watermarks, which Google describes as markers “imperceptible to the human ear,” allowing listeners and platforms to identify whether music was generated or edited by AI. In terms of copyright guardrails, the API documentation explicitly states: all prompts pass through safety filters, and requests targeting specific singers’ voices or generating copyrighted lyrics will be blocked.
In terms of musician involvement, the Lyria series had producers and musicians participate during development, and it listed several musicians who have used Google music tools: producer Wyclef Jean used Music AI Sandbox, while Yung Spielburg used Lyria to score the animated short film “Dear Upstairs Neighbors.”
Lyria 3.5 is now available across four entry points: Gemini App, Google Flow Music, Google AI Studio, and Google Vids. For general users, Gemini App is the most direct entry point; for developers, the API provides more granular control; for music creators, Flow Music remains the dedicated workflow. By widening these access points, Google is effectively advancing AI music generation from the level of “tools” to that of “conversation.”
Conclusion
The technical upgrade in bringing Lyria 3.5 to Gemini is limited: 44.1kHz, full song structure, SynthID watermarking, and single-turn generation all already existed during the Flow Music stage. The real change is in distribution—moving from Flow Music’s music creator community to Gemini, a user-facing entry point on the scale of hundreds of millions. Features like templates, short versus full-length song selection, and style options may seem unremarkable, but they’re designed to lower the prompt barrier, turning “generate a birthday song in Gemini” into a natural action that directly produces a song.
Google is pushing Lyria 3.5 to the biggest entry point now, both to capture user habits early and to stake out its position with SynthID and guardrails while copyright rules are still murky. For users who are already used to asking questions, organizing documents, and writing code in Gemini, the distance to generating a song has shrunk from “opening another app” to “typing one prompt.”
Source: KOCPC Chinese