GPT Image 2 Prompt Generator

Problem

Writing prompts for GPT Image 2 is less about sentence style and more about translating requirements into model-friendly input. A Chinese description may mix subject, style, lighting, composition, and use case, while a direct prompt can leave out the main subject, place style tags in the wrong position, or miss concrete camera and art terms. This skill breaks that conversion into inspectable steps and produces an English prompt structured for image generation.

How it works

The skill works with several signals:

• Requirement parsing: identify the subject, style preference, technical parameters, and intended use, such as a poster, product shot, or character scene.
• Template-based generation: use references/prompt-templates.md, references/style-guide.md, and references/technical-terms.md to organize the prompt into subject, style, environment, lighting, and composition.
• Terminology enrichment: add photographic, artistic, and compositional terms such as backlight, soft light, top-down view, close-up, rule of thirds, and depth of field.
• Negative prompting: derive exclusion constraints from requests to avoid certain elements, reducing common issues such as malformed hands, garbled text, or unstable proportions.

The output usually includes a ready-to-use English prompt, optional negative prompts, and parameter notes explaining which style and technical choices were applied.

Boundaries

It fits concrete image-brief tasks such as posters, photography, characters, scenes, and logos. Very abstract requests like premium feel or good atmosphere still need concrete subjects, materials, lighting, and usage details. It does not generate images directly and does not guarantee identical results across platforms or model versions, so the prompt should still be refined against actual outputs.