Google DeepMind music-generation model for high-fidelity tracks with vocals, lyrics, detailed prompting and variable duration up to three minutes, available through Gemini/Google creative surfaces and downstream API distribution.
Lyria 3.5 is Google DeepMind's current full-length music-generation model. It generates structured songs from text prompts, including vocals and lyrics, and current Google surfaces also support image-conditioned composition. DeepMind positions it as its newest Lyria model with improved vocals, musicality and variable song lengths up to three minutes.12
For developers, the Gemini API exposes model `lyria-3.5` through `generateContent` and the Interactions API. The current developer documentation specifies 44.1 kHz stereo MP3 output, text and image inputs, structured sections such as verses, choruses and bridges, and prompt-guided duration. Music generation is single-turn: iterative editing of an already generated track is not supported in the current API.3
Access is broad. Google made Lyria 3.5 available globally in the Gemini app, in Google Flow Music, Google AI Studio, Google Vids and through the Gemini API.43 That gives it both consumer and developer entry points rather than limiting it to a specialist creation product.
The exact current first-party numeric API price for the `lyria-3.5` model was not verified on the reviewed Google pricing surface, which still exposes pricing for earlier Lyria 3 preview identifiers. AiToolMap therefore does not import third-party price claims into the canonical review. Consumer access exists through Google apps subject to their own limits and plan terms.
Google publishes a dedicated Lyria 3.5 model card describing its architecture, training-data processing, evaluation categories, intended use and safety mitigations. Google reports qualitative gains over Lyria 2 in audio fidelity and lyric prompt adherence, but these are first-party evaluations and no vendor score is converted into independent rating evidence.2
All generated audio is watermarked with SynthID. Google also documents safety filtering, including blocking requests for specific artist voices or copyrighted lyrics in the developer surface.13 The Gemini API terms state that Google does not claim ownership over generated content, while users remain responsible for lawful use and any applicable rights obligations.5
The fixed 50-source panel found no substantive independent exact-model review or eligible numeric rating. The Apple App Store listing for Google Flow Music does explicitly identify Lyria 3.5 as the current model and corroborates current feature availability, but the listing lacks sufficient review volume and represents the host app rather than a clean model-quality benchmark. It is therefore context only, not a numeric input.6
Lyria 3.5 is best suited to creators, video teams, marketers and developers who want complete songs or soundtracks from text or visual concepts without running their own audio model. It is particularly attractive when work already happens inside Gemini, Flow Music, Vids or a Google AI Studio/API stack.43
It is less suitable when a workflow depends on iterative section-level editing inside the API, audio-reference conditioning, voice cloning or prompts designed to imitate a named artist. Those are either unsupported or deliberately restricted in the current surface.3
the combination of full-song generation, vocals and lyrics, high-fidelity stereo output, image-conditioned creation, duration/structure prompting, broad Google distribution and explicit provenance watermarking.13 Google also provides more model-card transparency than many closed music generators.2
mainly evidence and workflow boundaries. Independent exact-model testing is still sparse, current API generation is single-turn, and exact 3.5-specific API pricing was not verified on the first-party pricing surface. The model is therefore scored on documented current capabilities and access, with LOW confidence because external product-quality validation remains thin.
Products matched by shared tasks, audience, workflow, product type and capabilities.