MODEL SIGNAL
Lyria 3.5
Google brings high-fidelity, full-track music generation to the Gemini API, blending text and image prompting into structured stereo audio.
Bottom line
Google has released Lyria 3.5, its flagship music generation model capable of producing 44.1 kHz stereo audio—complete with verses, choruses, bridges, and vocals—from text and image prompts. Launched on July 29, 2026, the model is available to developers via standard Gemini API endpoints and to consumers through Google Flow Music and the Gemini app.
Signal
The primary signal here is the formal integration of production-grade audio generation into Google's standard developer tooling. Lyria 3.5 isn't isolated in a bespoke audio sandbox; it sits within the Gemini API (Interactions and generateContent APIs). By supporting multimodal inputs—specifically allowing image prompts to influence audio generation—and natively returning both audio files and lyrics text, Google is treating music as a first-class programmatic output. The directional implication is that developers can now chain visual and textual contexts directly into structured, full-length musical compositions without switching ecosystems.
Noise
The 131,072 context window is listed as a hard spec, but for operators building music applications, raw text-token context translates opaquely to audio duration and generation latency. Until operators stress-test the model's memory across multi-turn musical iterations, treat the context window as a technical ceiling rather than a practical measure of audio-generation length. Additionally, claims that it is Google's "best-sounding" AI music model are expected provider marketing and should be evaluated against your specific production requirements.
Model profile
Lyria 3.5 (model ID: lyria-3.5) is a multimodal music generation model. It accepts text and image inputs and outputs both audio and text (lyrics). The model generates 44.1 kHz stereo audio, producing MP3 files by default with optional WAV output via standard response_format settings. It is capable of generating full-length songs with structured components, including verses, choruses, and bridges.
Assessment
Google asserts that Lyria 3.5 offers improved musicality, enhanced lyrics, and more expressive vocals compared to earlier iterations of the Lyria family. The operator read is that Google has crossed the threshold from generating short, ambient soundscapes to generating structured, lyrical tracks. By outputting both the high-fidelity audio and the accompanying lyrics text, it provides a comprehensive asset package for end-user applications.
Where it fits
Lyria 3.5 fits squarely into creative tooling, dynamic media generation, and interactive entertainment. Game developers can use image prompts of scenes to dynamically generate bespoke soundtracks. Content platforms can offer users automated scoring that includes actual vocals and song structure. It is also an obvious fit for AI-assisted music production workflows where rapid prototyping of lyrics and melodies is required.
Operator implications
Developers routing to lyria-3.5 will need to handle dual-output payloads, managing both the generated audio artifact and the returned lyrics text. Operators should configure their response_format explicitly based on their pipeline's latency and fidelity needs—defaulting to MP3 for web-streaming applications or opting for WAV when feeding the audio into downstream mixing or mastering pipelines.