Google DeepMind launched Lyria 3.5 in Google Flow Music on Wednesday, delivering its most expressive AI music generation model yet — with meaningfully better vocals, sharper lyric coherence, and direct tempo control — while its two main rivals remain mired in copyright litigation that Google has deliberately structured itself to avoid.

The update is live now for all Google Flow Music users at no additional cost. It is the third significant Lyria milestone since February, a pace that signals Google is treating AI music not as a research experiment but as a priority product line, according to Google’s official announcement.

What Lyria 3.5 Changes

According to Google’s official announcement, the update touches four distinct areas.

On musicality, the model now generates richer and more complex melodic structures that sound more natural across the length of a track. This directly addresses a persistent criticism of earlier Lyria releases: melodic loops that felt mechanical or repetitive after the first 30 seconds.

On lyrics, Lyria 3.5 produces higher-quality text with what Google describes as improved prompt adherence and structural awareness. In practice, this means the model is more likely to write lyrics that actually fit the mood, narrative, and genre you described — rather than producing grammatically correct but thematically hollow text.

On vocals, the update brings more realistic and emotionally nuanced performances, along with improved pronunciation. This matters particularly for creators working with non-English lyrics, genre-specific vocal styles, or any content where expressive delivery is the point rather than generic delivery that merely hits the correct pitch.

On creative control, users can now set tempo and duration with more precision — a practical upgrade for producers who have found earlier Lyria outputs useful as sketch material but too unpredictable for structured compositions.

How AI Music Generation Actually Works — and Why Lyria Sounds Different

Understanding what separates Lyria from its competitors requires understanding the underlying technology.

Lyria 3.5 — like its Lyria 3 and Lyria 3 Pro predecessors — outputs 48kHz, 16-bit PCM stereo audio. That is professional production quality. The underlying stack is not a single model: Lyria handles audio generation, Gemini provides the natural language understanding that interprets your creative prompt, and Veo handles any visual components in the broader Flow ecosystem, as explained in a comprehensive guide to Google Flow Music and Lyria 3 Pro. ProducerAI, the startup Google acquired in February 2026, was built on a different architecture — the original Riffusion model, which fine-tuned Stable Diffusion on spectrograms, treating music as an image generation problem. Google replaced that architecture entirely with Lyria 3 at acquisition.

The structural coherence improvements that debuted in Lyria 3 Pro — and that Lyria 3.5 now builds on — come from training the model to understand songs as structured compositions with intros, verses, choruses, and bridges, rather than as continuous audio to generate token by token without knowledge of where it stands in a song’s arc.

Where Lyria 3.5’s vocal improvements come from is harder to pin down without a technical paper, but independent comparisons placed Lyria 3 Pro “comparable to Suno v4.5 in quality” on overall audio fidelity — not yet at Suno v5’s level, but closing the gap. The Lyria 3.5 release targets exactly the vocal territory where Udio has historically been strongest.

What SynthID Means Beyond Watermarking

Every track generated by Lyria carries a SynthID watermark — but SynthID has become something larger than a single product’s feature.

Google’s audio watermarking deploys psychoacoustic masking: an imperceptible cryptographic signal is embedded directly into the audio waveform at generation time, exploiting frequency ranges below human hearing sensitivity and the masking effect during louder segments. The result is a digital signature that is part of the sound itself, not a removable metadata tag. It survives MP3 compression (including aggressive 128kbps encoding), speed and pitch shifts, and even re-recording through a speaker.

In May 2026, OpenAI adopted SynthID watermarking for images generated by ChatGPT, and NVIDIA integrated it into its Cosmos world models, as documented at the SynthID cross-industry tracking resource. SynthID is becoming the default AI content provenance standard across the industry — which means Lyria-generated music sits inside the only verifiable chain of custody that streaming platforms, rights management organizations, and regulators can use for automated detection and policy enforcement. Deezer’s 99.8% AI-detection system, Spotify’s AI Credits disclosure framework, and TIDAL’s July 15 AI-labeling policy all depend on being able to identify AI-generated audio. SynthID is how Google’s output gets that label reliably.

Suno and Udio do not use SynthID.

Lyria’s Development Cadence

Wednesday’s release is the third significant Lyria milestone in five months:

Lyria 3 launched February 18, 2026, generating 30-second high-fidelity tracks from text or image prompts through the Gemini app — for users 18 and older — at 48kHz with lyrics, vocals, and custom cover art, as reported by Android Central.

Lyria 3 Pro followed March 25, 2026, extending the maximum track length to three minutes and introducing the structural song awareness that distinguishes proper full-length compositions from extended loops. Google rolled Lyria 3 Pro out across six platforms simultaneously: the Gemini app, Google Vids, ProducerAI (now Flow Music), Vertex AI, the Gemini API, and AI Studio, per the complete Flow Music guide.

Lyria 3.5 now builds on the 3 Pro foundation, focusing on expressive quality rather than duration or distribution breadth.

Google Flow Music: What the Platform Has Become

Google Flow Music — the platform where Lyria 3.5 lives — is the renamed and rebuilt version of ProducerAI, itself formerly known as Riffusion, which Google brought under its Google Labs umbrella in February 2026. The platform lets users generate songs from text, image, audio, and voice prompts; remix existing tracks; edit compositions at the section level; rewrite lyrics in place; publish creations; and collaborate through shareable workspaces, as detailed in the complete Flow Music guide.

In May 2026, Google partnered with Believe — the global artist development company that distributes for TuneCore — to offer Flow Music directly to Believe’s artists and songwriters. The partnership marked a deliberate shift: Google is not just building a consumer toy, it is positioning Flow Music as infrastructure for professional and semi-professional music creation.

Flow Music also integrates with Google Vids, Google AI Studio, Vertex AI, and the Gemini app — a distribution footprint neither Suno nor Udio can match.

Why Licensed Training Data Is Now a Structural Advantage

The clearest differentiator Google presses is its training data provenance. Lyria 3 Pro — and by extension Lyria 3.5 — is trained on licensed data using materials that YouTube and Google have the right to use under their terms of service, partner agreements, and applicable law. Google confirmed the model does not replicate any specific artist’s voice or catalog.

That matters enormously in 2026. The Recording Industry Association of America filed landmark copyright lawsuits against both Suno and Udio in June 2024 on behalf of Sony Music, Universal Music Group, and Warner Music Group, according to AI Vortex’s lawsuit status tracker. Universal Music Group settled with Udio in October 2025. Warner Music Group settled with Suno in November 2025. But Sony Music Entertainment has settled with neither — and its cases against both companies remain fully active. Suno faces a maximum statutory damages exposure that could reach into the billions of dollars as Sony and Universal moved to expand their complaint to 61,026 specifically identified recordings, as detailed in TechTimes’ coverage of the lawsuit.

In July 2026, hacked Suno source code confirmed that the company used a proxy tool to stream-rip audio from YouTube Music, Deezer, and other streaming platforms for training — a practice the company did not deny and that its legal team has argued is not prohibited by the DMCA, as reported by TechTimes.

For commercial creators, brands, sync licensing operations, and enterprise workflows, this litigation cloud is a real risk that Google’s licensed-data approach eliminates entirely.

How Lyria 3.5 Fits a Competitive Field

Suno remains the market leader by every quantitative metric: 2 million paid subscribers and $300 million in annual recurring revenue as of February 2026, with over 100 million people having used the platform since launch, per Music Business Worldwide — and a $5.4 billion Series D valuation as of June 2026, according to Sacra. Independent community benchmarks placed Suno v5 at an ELO score of 1,293 — ahead of every competitor on audio fidelity, musical structure, and vocal realism as of early 2026.

Udio, built by former Google DeepMind researchers, has cultivated a loyal following among electronic music producers for its tight bass response, instrument-aware generation, and stem separation capabilities. Community comparisons in early 2026 gave Udio a quality edge over Suno specifically on electronic, hip-hop, and pop — the genres where vocal timing and production crispness matter most.

Google’s advantage going into Lyria 3.5 was distribution breadth, not audio quality leadership. The vocal and lyric improvements in today’s release are a direct attempt to close the gap with Udio’s vocal territory while maintaining Google’s structural advantages: the legal cleanliness of licensed training data, SynthID watermarking that survives common audio transformations, and integration with the Gemini ecosystem that Suno and Udio cannot replicate.

What Does This Mean for Creators?

For musicians and content creators already working inside Google Flow Music, Lyria 3.5 is a free upgrade with no additional subscription tier and no waitlist. The tempo and duration controls are the practical near-term gains: producers who have found earlier Lyria outputs useful as sketch material but too unpredictable for structured compositions now have more direct control over the raw material.

The more significant long-term factor is structural. As streaming platforms implement mandatory AI-content labeling — Spotify’s AI Credits disclosure launched April 16, 2026; TIDAL’s AI policy took effect July 15, 2026 — the ability to trace a track’s provenance becomes commercially important. SynthID gives Lyria-generated music a verifiable identity that every other major AI music generator currently lacks or has not standardized.

Whether the vocal improvements in Lyria 3.5 are sufficient to pull producers away from Suno v5’s established workflows is a question three months of actual output will answer more reliably than any launch day claim. But with three major releases in five months, a platform that has expanded from a music lab into a full creative studio, and a competitive landscape where its two main rivals face litigation that could reshape what AI music is even legal to build on — Google DeepMind has positioned Lyria 3.5 as the version where the gap between “safer” and “better” begins to close.

Frequently Asked QuestionsWhat is new in Lyria 3.5 compared to Lyria 3 Pro?

Lyria 3.5 focuses on four specific improvements over the March 2026 Lyria 3 Pro release: richer and more naturally flowing melodic structures, higher-quality lyrics with better prompt adherence and structural awareness (meaning the lyrics actually fit what you described, not just what sounds grammatically plausible), more emotionally expressive and pronounced vocals, and direct controls for setting the tempo and duration of generated tracks. Lyria 3 Pro introduced structural song coherence — intros, verses, choruses, bridges — and the three-minute duration ceiling. Lyria 3.5 builds on that foundation and makes the output sound more like it was performed rather than generated.

Is Google Flow Music free, and does Lyria 3.5 cost extra?

Lyria 3.5 is available inside Google Flow Music at no additional cost for existing users. Google Flow Music operates on tiered access that maps to Google AI subscription plans: Google AI Plus members get the Starter plan’s benefits, AI Pro members get the Plus plan, and AI Ultra members get the Member plan. There is no separate paywall or waitlist for the Lyria 3.5 upgrade.

How does Lyria’s copyright status differ from Suno and Udio?

Lyria is trained on licensed music — data that YouTube and Google are contractually authorized to use under partner agreements and terms of service. Suno and Udio were both sued by major record labels in June 2024 for training on copyrighted recordings without permission. Universal Music Group settled with Udio in October 2025; Warner Music Group settled with Suno in November 2025; Sony Music Entertainment has settled with neither and remains in active litigation against both. For commercial creators, brands, and sync licensing work, Lyria-generated tracks come without the legal exposure that still surrounds Suno and Udio output.

What is SynthID and why does it matter for AI music?

SynthID is Google DeepMind’s audio watermarking technology. It embeds a cryptographic signature directly into the sound’s waveform at generation time — not in a metadata tag that can be stripped — using psychoacoustic masking to keep the signal below the threshold of human hearing. The watermark survives MP3 compression, speed and pitch changes, and re-recording through a speaker. As streaming platforms like Spotify, TIDAL, and Deezer roll out mandatory AI-content disclosure systems in 2026, SynthID gives Lyria-generated tracks a verifiable chain of origin that automated detection systems can confirm. In May 2026, OpenAI adopted SynthID for ChatGPT-generated images, and NVIDIA integrated it into its Cosmos models — meaning SynthID is becoming the AI content provenance standard beyond Google’s own ecosystem.