Google: Lyria 3 Pro Preview

Generates full-length songs (typically 1-3 minutes) from text prompts and optional lyrical input, using Google's proprietary diffusion-based music synthesis architecture trained on licensed music data. The model accepts natural language descriptions of musical style, mood, instrumentation, and tempo, then synthesizes coherent audio at 48kHz sample rate with maintained harmonic structure across the generated duration. Integration occurs via REST API calls to the Gemini API endpoint with async job polling for generation completion.

Accepts high-level semantic descriptions (genre, mood, instrumentation, cultural style, tempo range) and translates them into latent music representations via a learned prompt encoder, then synthesizes audio that matches the specified aesthetic without requiring technical music notation or MIDI input. The model uses a two-stage pipeline: semantic understanding via transformer-based prompt encoding, followed by diffusion-based audio synthesis conditioned on the encoded representation. Supports natural language variations like 'upbeat indie pop with lo-fi production' or 'melancholic orchestral with strings and piano'.

Provides asynchronous API endpoints for submitting music generation requests and polling for completion status, enabling non-blocking workflows where generation jobs run server-side while client applications continue execution. Implements standard async patterns: request submission returns a job ID, client polls a status endpoint at intervals, and completed generations are retrieved via a results endpoint. Supports batch submission of multiple generation requests with individual job tracking, enabling pipeline parallelization and cost-aware scheduling.

Accepts user-provided lyrics or lyrical themes and generates music that maintains semantic and emotional alignment with the text content, using a joint embedding space that encodes both lyrical meaning and musical characteristics. The model conditions the diffusion process on lyrical embeddings, ensuring generated melodies and harmonies reflect the emotional arc and narrative of the lyrics. Supports partial lyrics (chorus only, verse structure) or full song lyrics, with the model inferring musical phrasing and cadence to match lyrical structure.

Exposes music generation capabilities through standard REST endpoints compatible with the Google Gemini API ecosystem, enabling integration with existing Google Cloud workflows, authentication systems, and monitoring infrastructure. Requests are authenticated via OAuth 2.0 or API key, with responses following Gemini API conventions for error handling, rate limiting, and metadata. Supports standard HTTP methods (POST for generation, GET for status) with JSON request/response bodies, enabling integration with any HTTP client or SDK.

Generates audio at 48kHz sample rate (professional studio standard) using diffusion-based synthesis that produces perceptually high-quality output with minimal artifacts, noise, or distortion. The synthesis pipeline operates in the frequency domain or learned latent space to maintain audio coherence across long durations (1-3 minutes), with post-processing to ensure smooth transitions and consistent loudness levels. Output is suitable for professional music production, streaming platforms, and broadcast without additional mastering or enhancement.