Google DeepMind's Lyria 3.5 Takes Direct Aim at Suno's Vocal Dominance

Google DeepMind has released Lyria 3.5, its latest AI music generation model, deploying it directly inside Google Flow Music. The upgrade touches every layer of the stack: prompt adherence, vocal expressiveness, melodic transitions, and precise control over tempo and song length.

From research toy to production tool

The Lyria family has moved fast. Lyria 3 launched with 30-second high-fidelity tracks generated from text or image prompts through the Gemini app. Lyria 3 Pro expanded that to full songs up to three minutes, adding structural coherence so tracks could include proper intros, verses, choruses, and bridges. Lyria 3.5 pushes the model further, prioritizing quality over length.

Google Flow Music is an AI-driven platform for composing, publishing, and sharing original music. Its "Chat with Producer" interface generates full-length songs with dynamic vocals, and the platform draws on Gemini, Veo, and Nano Banana models for chat, video, and image generation. Think of it less as a Suno clone and more as an AI-native DAW being assembled piece by piece.

What actually changed in 3.5

The four pillars of the upgrade, straight from the official announcement:

• Improved musicality: Richer, more complex melodic structures that sound more natural.
• Enhanced lyrics: Higher quality lyrics with better prompt adherence and structural awareness.
• Improved vocals: More expressive, emotionally nuanced singing with improved pronunciation.
• Creative control: Exact BPM targeting and stem exports for full-length songs.

BPM control and stem exports are the most practically significant additions for producers. Previously, AI music tools largely delivered a finished stereo file with no way to pull it apart. Flow Music now offers remix, audio effects, and stem separation to isolate vocals, drums, bass, and melody. That means AI-generated components can drop directly into a traditional DAW workflow rather than arriving as a take-it-or-leave-it deliverable.

The competitive landscape

Suno and Udio have been in this space longer. On pure audio fidelity, Lyria 3 is considered the most technically impressive of the leading models, producing instrumental tracks with dynamic range and timbral realism that competitors haven't fully matched. Vocals have historically been where Suno pulls ahead. Lyria 3.5's explicit focus on more expressive, emotionally nuanced singing targets that gap directly.

Pricing keeps the barrier low: a 30-second Lyria 3 clip via API costs around $0.04, and a full three-minute song via Lyria 3 Pro runs roughly $0.08. Commissioning basic stock music from a human composer typically costs hundreds of dollars.

Google's bigger play

The model upgrade is one piece of a larger strategy. Believe, the global music distribution company, has partnered with Google to offer Flow Music to artists across Believe and TuneCore. The logic: get the next generation of musicians using AI as part of their creative process from the start, betting that early adoption by emerging artists will shape long-term platform loyalty.

As part of the deal, Believe and TuneCore will select a group of artists and producers to meet weekly with the Flow Music product team, feeding back directly into the platform's development. The arrangement suggests Google wants Flow Music embedded in distributor and label workflows, not just used as a hobbyist tool.

The copyright shadow

A group of independent musicians has filed a lawsuit against Google alleging it trained Lyria 3 on copyrighted YouTube recordings without permission or compensation. Google says Lyria 3 Pro was developed using materials that YouTube and Google have the right to use under their terms of service, partner agreements, and applicable law. That legal question remains open.

On ownership, Google says creators retain rights to original content produced with Flow Music. AI-generated outputs carry SynthID watermarking, Google's system that embeds an invisible signal directly into the audio waveform. The watermark survives compression and re-encoding, so generated content can be identified even after post-processing.

What this means for your workflow

• Lyria 3.5 is live now in Flow Music with no model switch required.
• Exact BPM targeting in prompts is a meaningful upgrade for sync licensing and beat-matching.
• Stem exports let you pull AI-generated tracks into any DAW for further editing.
• A free plan with daily credits lets you test the new model without a subscription.
• Fully synced music videos and other advanced multimodal features require an active Google AI subscription.

Three meaningful upgrades in under six months signals that Google is treating this as a priority product. Whether Lyria 3.5 closes the vocal quality gap with Suno is something the community will stress-test quickly, but precise tempo control and stem exports represent a genuine step toward professional utility.