DRULES AI
Google has rolled out Lyria 3.5 across the Gemini app, API, and related tools, letting users generate complete, structured songs up to three minutes long instead of short loops.
🎵 New Capabilities Unlock Pro Workflows
Users can now start from text prompts, genre selectors, or upload up to 10 images that the model analyzes for mood, color, and vibe before composing a matching soundtrack. Options include instrumental or full vocal tracks, custom lyrics, precise tempo/BPM control, and detailed section shaping for intros, verses, choruses, bridges, and outros. Output arrives as 44.1kHz stereo in MP3 or WAV, complete with an invisible SynthID watermark that flags it as AI-generated.
The model improves on earlier versions with richer arrangements, more emotionally nuanced and better-pronounced vocals, stronger prompt adherence, and more natural melodic development. Templates for background music, jingles, birthday tracks, or video scores speed up ideation. Pricing through the API sits at $0.08 per full track and $0.04 for 30-second clips, making it accessible for developers embedding music into video editors or games.
🔬 How It Stacks Up Against Dedicated Platforms
This move puts Google in direct competition with Suno and Udio. While those platforms focus exclusively on music, Lyria 3.5 lives inside Google's broader creative suite including Google Vids, AI Studio, and Flow Music. Seamless handoff from music generation to video syncing gives creators an end-to-end pipeline that previously required multiple tools and exports. Early tests shared on X show strong coherence across full song structures, though some users note occasional generic phrasing in lyrics that still requires human editing for professional releases.
Guardrails block voice cloning of specific artists and copyrighted lyrics, reflecting lessons from recent European court scrutiny of rival services. The watermarking initiative helps platforms and rights holders identify AI content at scale, potentially easing some industry fears around flooding catalogs with undetectable tracks.
📈 Implications for Working Creators
Professional musicians and content creators gain fast prototyping for sync licensing, social content, podcasts, and game audio. A video editor can generate a custom underscore from a rough cut screenshot, iterate tempo to match cuts, then export watermarked masters. Indie artists experimenting with AI can layer Lyria stems with live vocals or instruments to stay on the right side of emerging eligibility rules from bodies like ARIA.
Limitations remain: generated tracks cannot be edited mid-prompt, and heavy reliance without human input risks sounding formulaic. Yet the low barrier and high fidelity lower the cost of experimentation dramatically. Integration with Gemini's existing multimodal features suggests future updates could allow real-time refinement using voice commands or follow-up prompts like "make the chorus more anthemic."
Early community feedback on X highlights excitement around image-to-music for visual artists and filmmakers who previously struggled to license or commission fitting scores quickly. With global availability on web and mobile, this release accelerates AI music from niche experiment to daily utility.
Bottom line: Google just commoditized full-song AI generation inside tools millions already use, shifting the competitive edge toward ecosystem integration and responsible labeling.