All groups

Say what the image is for, then the visible details, then the exact text in quotes with position, then a short list of what must not appear.

GPT Image 2.5

Clean designed images with correct text.

Prompt instructions

A photograph of [your subject] in [your setting]. Exclusions: no extra text, no watermark, no logo.

GPT Image 2.5 reads photography language, exclusions allowed.

SubjectSetting

Unlike Nano Banana, exclusions work here when they are short and grouped.vendor: OpenAI image prompting guide via Director Chair field notes on image prompting
A city name gave the wrong city. Describe the materials and the street instead.measured: Director Chair field notes on image prompting

What the image is for: a product photograph, an advertisement, a diagram.
The visible details: materials, light, colours, medium, scale.
People: framing of the body, relative size, where they look, what they do with objects.
Text: the exact words in quotes, with position and type; unusual words and brand names spelled letter by letter.
Exclusions, short and grouped: no extra text, no watermark, no logo.

Change one thing per round and repeat the list of what must stay (identity, layout, light, labels) every round; details drift across repeated edits.

A city name drew the wrong city's architecture; name the materials instead (brick with blue stone).

Picks the image model by the job instead of by the hype of the week.

Stills that hold the reference and do not look generated.

Own words from Director Chair field notes on image prompting Checked 2026-10-13.