gpt-image-still-prompt (still-image prompt architecture)

Turns a one-line shot intent into a tight, photographer-brief prompt for gpt-image-2. The still counterpart to higgsfield-video-prompt. Use it every time we generate a still, so image prompts follow one repeatable structure instead of a keyword salad.

Where this sits (don't duplicate siblings)

Skill	Layer	Question
`static-ad` / `vid-ad`	Creative brain	What still/ad should exist, concept, range?
gpt-image-still-prompt (this)	Prompt engineering (STILL)	What words make gpt-image render this frame photoreal + consistent?
`higgsfield-video-prompt`	Prompt engineering (VIDEO)	(the animation sibling, motion, not look)
`static-qa`	QA gate	Did the still come out good / defect-free?

The #1 idea: prompt gpt-image like a creative brief for a photographer, not a tag list. Every clause is a visual FACT (a real lens, a real light, a real material), never a vibe word.

The architecture (ordered, write in this order, line-break between blocks)

Rich and specific, but every clause is a visual fact. Refine with small single-change follow-ups, not one giant prompt.