Image prompting

Translate the user's idea into a prompt that describes the desired image and protects the details they care about. This is a prompt-writing skill for Codex, Claude Code, and other compatible agents. It does not assume a particular image-generation provider or start image generation by itself.

For GPT Image 2.5, read references/gpt-image-2-5.md. For another specified image model, preserve the useful creative brief but adapt syntax and settings only to that model's documented controls.

Draft with intent

Identify whether the user wants to create an image or edit one. For an edit, distinguish:

• What should change.
• What must stay the same, such as identity, product shape, layout, camera angle, or palette.
• Which reference image controls each attribute.

Capture the few decisions that materially affect the result: intended use, subject, composition, aspect ratio or crop, visual style, exact text, and any preservation constraints. Ask only for missing essentials. If the request is clear enough, state reasonable assumptions and draft.

Describe the image in priority order: