DRULES AI
Google DeepMind launched Lyria 3.5 on July 29, immediately rolling it out in Flow Music with major upgrades to musicality, lyrics, vocals and creative controls. The model produces more natural melodies, coherent lyrics and expressive, dynamically sung vocals while adding pro-level features that address longstanding workflow complaints in AI music tools.
🎤 Vocals, Musicality & Core Model Upgrades
Lyria 3.5 generates richer arrangements that flow organically instead of feeling mechanically assembled. Vocals now convey genuine emotional range with clearer pronunciation and dynamic phrasing that rivals human performances in early demos. Lyrics show improved thematic consistency and rhyme quality, reducing the nonsensical outputs that plagued earlier versions. Tracks can run from 30 seconds to three full minutes with precise tempo locking that keeps everything grid-ready for DAW sessions.
One producer generated a 2000s hip-hop podcast intro from a single-sentence prompt, calling the output "insane" quality. Community tests confirm the model follows creative direction far more reliably, producing full-length songs instead of 30-second loops. Flow Music also added cover versions, lip-sync video generation and a new iOS app, expanding beyond desktop experimentation.
🎛️ Selective Editing and Stem Export Revolution
The headline feature is Selective Section Painting, letting users isolate and regenerate specific parts of a track without starting from scratch. Need a better bridge or different vocal tone in verse two? Paint the section, describe the change, and iterate. Combined with exportable stems for vocals, drums, bass and instrumentation, this moves AI music from novelty to production-ready asset.
Early feedback indicates stem separation is clean enough for professional mixes, potentially disrupting large segments of the sample library market overnight. Producers can now generate, stem out, tweak in Ableton or Logic, and blend with live instruments or original MIDI. BPM control ensures instant sync with existing projects, eliminating the tedious realignment work that defined prior tools.
Unlike Suno and Udio, which still rely on unofficial API wrappers that risk account bans, Google is offering more official pathways. Integration with Spaces "vibe coding" tool further embeds music generation into Google's creative ecosystem. The iOS app lowers the barrier for mobile ideation, letting beatmakers sketch during commutes or travel.
📈 Ecosystem Impact and Remaining Gaps
This release arrives as the AI music sector grapples with legal uncertainty elsewhere. Google's focus on usability and control could attract professionals frustrated with "take-it-or-leave-it" generation. However, even advanced models struggle to consistently birth viral hits without significant human refinement. Skeptics note the output still requires curation, arrangement tweaks and final mastering to compete commercially.
Expect a wave of tutorials on optimal prompting for section painting and stem workflows in coming days. Early adopters experimenting with the new controls are already sharing genre-spanning results, from Polynesian mixes to orchestral pieces. As competitors respond, the entire category benefits from raised standards around controllability and integration.
Bottom line: Lyria 3.5 finally delivers the editing precision and stems needed to make AI a true production partner instead of a party trick.