GPT Image Prompting Guide

What problem it solves

Image prompting often breaks down when generation, editing, multi-image references, and exact text are mixed into one prompt. gpt-image-prompting-guide turns Chinese or natural-language requests into an English prompt that can be pasted into ChatGPT's web UI. It focuses on preserving subject, composition, lighting, text, required changes, and protected details instead of calling an image-generation API.

How it works

• Mode selection: uses the web-prompt path for browser, copy, paste, or unspecified requests; routes explicit API, SDK, or code requests to an API variant.
• Requirement extraction: keeps brand names, numbers, proper nouns, exact copy, reference-image roles, edit targets, and immutable features.
• Template composition: selects the closest scenario in resources/prompt-templates.md and stacks at most adjacent methods to avoid conflicting styles.
• Output structure: orders the English prompt by goal, subject, composition, text, and constraints; editing prompts state Change, Preserve, and Do not change; multi-image prompts use Image 1, Image 2 labels.

Boundaries

It returns a paste-ready prompt, a Chinese translation, and key assumptions only. It does not generate images, return code, SDK calls, model parameters, or promise one-shot success. It also does not provide ways to bypass safety limits.