Google expands Lyria 3.5 to Gemini and the Gemini API worldwide

Google has broadened access to Lyria 3.5, its current flagship music-generation model. The model first appeared in Google Flow Music in July, but on September 4 Google announced global availability inside the Gemini app and a developer surface through the Gemini API and AI Studio.\n\nFor ordinary Gemini users, the expansion makes music generation a first-party Gemini workflow rather than a separate Labs experience. Google says users can choose or describe a genre, select vocal or instrumental output, start from templates and ask for different track lengths. DeepMind’s current Lyria page describes the model as supporting richer vocals, musical arrangements and tracks up to roughly three minutes.\n\nThe developer surface is the more important change for the AI-tool ecosystem. Google’s current API documentation lists Lyria 3.5 for text/image-to-music generation and high-fidelity stereo output. The Gemini API pricing page currently lists no free tier for the full-song model and a paid price of $0.08 per song/request. That price should be treated as the current developer API tariff, not as the cost of Gemini consumer access.\n\nGoogle also documents limitations that matter before developers build workflows around it. Generated audio carries an imperceptible SynthID watermark. The current model does not support multi-turn music editing: each generation is a single-turn operation, so users cannot yet iteratively revise an existing track through a conversational sequence in the API. Results can also vary across repeated calls with the same prompt, and safety filters can block prompts involving prohibited content, specific artist voices or copyrighted lyrics.\n\nIndependent September 4 coverage from DataNorth likewise identified the API preview as a new access expansion rather than a new underlying model family. The model card itself predates this release and provides Google-run evaluation showing improvements over Lyria 2; AiToolMap does not treat those internal comparisons as neutral evidence of superiority over competing music models.\n\nFor the directory, Lyria 3.5 is now a stronger standalone model/tool candidate because it has a current developer API in addition to Google-owned creative interfaces. Canonical research should reconcile the exact Clip/Full Song model naming across the newly updated docs, current quotas and commercial terms, and whether independent creators can reproduce the quality and prompt-adherence gains Google reports.