Principles and Methods of Writing Prompts

My principles and methods when writing prompts are roughly as follows: My prompt principles 1. Make it a template, not a fixed prompt. This way everyone can freely adapt based on their own scenario - input their preferred brand, city weather, company stock, mini store theme, even turn articles into infographics. The core is giving users a playable framework, not dictating exactly what to write. 2. Fully leverage the model own capabilities (search, world knowledge, understanding). For example, city weather prompt lets Gemini retrieve weather based on city and date; mini 3D company stock chart also uses model real-time financial search; infographic prompt relies on model understanding and extraction of article content. My strategy for writing Nano Banana Pro 1. First use AI to run through a prototype for a specific situation. 2. Then abstract this prototype into an extensible prompt template, letting input content determine final output. The core is letting the model automatically combine structure, automatically adapt to scenarios, not artificially limiting it. Like writing programs, dont hardcode, but leave interfaces for flexible combination. About simplifying prompts: Current models are very capable, longer or shorter tokens dont matter much. So I wont spend excessive time compressing prompts, but prioritize ensuring it works and can grow. Consider simplifying when functions mature if necessary.