Google: Lyria 3 Pro Preview

Google: Lyria 3 Pro Preview is a music generation model from Google, part of the Lyria 3 family of music generation models. It is available through the Gemini API and is listed on api14 under the audio category, with text also noted as a related category. The model id is google/lyria-3-pro-preview, and it is a preview release from the provider.

The model accepts text and image inputs and can produce text and audio outputs. Its context window is 1,048,576 tokens, and the maximum output is 65,536 tokens. According to the provider description, Lyria 3 generates high-quality audio at 48kHz, and full-length songs are supported. That combination of a very large context window and audio output makes it suited to prompts that carry a lot of descriptive or structural detail, and to image-conditioned requests where a visual reference informs the generated music.

Practical uses follow directly from those capabilities. You can generate full-length songs from text descriptions, sketch musical ideas from an image, or produce audio alongside text in a single response. Because the input side accepts images as well as text, the model fits workflows where a mood board, artwork or reference still is meant to steer the resulting track. The large context window also means long, detailed briefs about arrangement, instrumentation or structure can be passed in without being trimmed.

You call the model through api14's OpenAI-compatible API using the model id google/lyria-3-pro-preview. The supported request parameters are max_tokens, response_format, seed, temperature and top_p. Because the interface is OpenAI-compatible, you can point an existing OpenAI client at api14 and set the model field to that id, then pass any of the supported parameters to control output length, response formatting, determinism via seed, and sampling behaviour through temperature and top_p.

It is worth noting that this is a preview model, so behaviour may change. It is also an audio generation model rather than a general text assistant, so if your task is ordinary text generation or reasoning without audio output, a text-focused model is likely the better fit. If your task is music or audio generation from text or image prompts, this model is the relevant choice.