Google Unveils Lyria 3.5 in Flow Music: Enhanced Control, Vocal Depth, and What It Means for Europe

What Is New in Lyria 3.5 and Google Flow Music

Google DeepMind's latest iteration of its audio architecture, Lyria 3.5, marks a significant shift from simple text-to-audio generation toward interactive musical composition. Built into Google Flow Music—Google's cloud-based audio creation platform—the model allows users to generate multi-minute tracks using detailed prompt conditioning, reference audio sketches, and structured lyric inputs.

Previous generations of AI music models often produced monolithic audio files where instruments, vocals, and sound effects were fused together. Lyria 3.5 addresses this limitation by introducing real-time stem generation, enabling creators to extract, isolate, and modify distinct tracks such as vocals, bass lines, drums, and harmonic accompaniment directly within the interface.

Key technical enhancements introduced in Lyria 3.5 include:

• Vocal Expressiveness and Polyphony: Enhanced natural pitch control, realistic vibrato, and multi-vocal harmonies, reducing the metallic artifacts common in older diffusion-based audio models.
• Acoustic Realism and Dynamic Range: Improved rendering of acoustic instruments, reverberation spaces, and dynamic changes from subtle quiet passages to heavy drop transitions.
• Multimodal Audio Conditioning: Users can input short voice hums, piano midi sketches, or rhythm taps as a primary guide alongside natural language descriptions.
• Precise Arrangement Control: Ability to define explicit track structures (intro, verse, chorus, bridge, outro) with custom tempo and key signature changes.

Competitive Landscape: Lyria 3.5 vs. Suno v4 vs. Udio v1.5

To evaluate where Lyria 3.5 stands in the current generative audio market, we compared its core specifications against leading industry competitors, Suno and Udio.

Pricing and European Availability

Google Flow Music powered by Lyria 3.5 is rolling out globally, including across all 27 European Union member states and the UK. Access is structured through Google's existing ecosystem tiers:

• Free Tier: Limited experimental credits in Google Labs for trial generation, restricted to standard quality mp3 exports with non-commercial usage terms.
• Google One AI Premium: Priced at €21.99 / $19.99 per month in Europe, providing expanded generation quotas, high-resolution 48 kHz uncompressed stem exports, and commercial usage rights within YouTube Creator Music.
• Enterprise & Cloud API: Available via Google Cloud Vertex AI, billed per audio generation minute (approximately €0.05 / $0.06 per minute of generated audio), aimed at game developers, media agencies, and software developers.

EU AI Act Compliance and SynthID Integration

One of the crucial aspects of the European launch is adherence to the regulatory landscape set by the European Union AI Act. Under Article 50 of the AI Act, AI model providers must ensure that synthetic audio, video, and text content are detectably labeled to prevent deceptive manipulation and uncredited copyright infringement.

Google DeepMind has integrated SynthID for Audio directly into the core synthesis pipeline of Lyria 3.5. Unlike surface-level metadata tags, which can be stripped away during audio conversion or editing, SynthID embeds an imperceptible digital watermark directly into the audio frequency spectrum. This watermark remains detectable by verification software even after audio compression, pitch shifting, or mixing into broader orchestral tracks.

For European media creators and broadcasting companies, this integrated watermarking simplifies compliance with local broadcasting standards and copyright disclosure rules without degrading the subjective listening experience.

Impact on European Creators and Copyright Licensing

The introduction of professional-grade tools like Google Flow Music presents both opportunities and legal questions for the European music sector. Licensing organizations across Europe (such as SACEM in France, GEMA in Germany, and JASRAC internationally) are actively evaluating how AI models train on copyrighted works and how revenues should be distributed.

Google has indicated that content created with Lyria 3.5 inside Google Flow Music can be cleared for commercial monetization on YouTube under specified agreement frameworks. However, tracks using generative elements must be flagged appropriately during upload in the YouTube Studio dashboard to comply with European consumer disclosure requirements.

For developers interested in evaluating open AI models or benchmarking performance, broader resources are available through our AI Arena benchmark breakdown, where hardware demands and latency metrics for local and cloud models are continuously tested.