Structured Prompt Guide: Generating a Five-Minute-Scale Score with Lyria 3 Pro

This piece turns "prompt engineering" itself into a listenable example: it uses lyria-3-pro to generate an instrumental score of about two minutes, with the prompt written in reusable template blocks—first genre and instrumentation, then mood and tempo, and finally timestamped section markers ([0:00 - 0:25] intro, etc.) to divide the structure and energy trajectory, with the mixing and stereo requirements stated at the end. The goal is not to make one specific song, but to give developers a skeleton they can adapt: explicitly writing out the four categories of genre, instrumentation, mood, and sections, so the model's understanding of structure is grounded in practice. The piece also submits an atmospheric reference image (image_urls), used to guide the overall tone rather than as a visual source, making it easy to compare the difference between the "text-only prompt" and "text + image guidance" approaches. Applicable scenarios include: pre-producing full-length background scores for videos or short films, producing instrumental material with a consistent style in batches for games and brand projects, and validating whether the prompt structure is stable before integrating with the Gemini API or Vertex AI.