Write image prompts that actually work

Image prompting is a distinct skill from chat prompting. The difference between a generic, stocky result and a striking one is almost always specificity — describing a scene the way an art director briefs a shoot rather than the way a customer orders a stock photo.

Subject and action; setting and background; lighting and mood; framing and aspect ratio. Mood and lighting in particular do most of the emotional heavy lifting, so never skip them. Leave a pillar out and the model defaults to bland.

Photography terms steer image models far more reliably than vague adjectives — 'shallow depth of field', 'soft diffused light', 'overhead flat-lay', 'rule of thirds'. Then iterate one element at a time rather than rewriting from scratch: change the light, then the angle, then a prop. Refining a near-miss is faster and more controllable than rolling the dice on a whole new prompt.

At the advanced level you're directing, not describing — specify where the subject sits, where negative space goes, the camera angle, and use a reference image when words can't capture the look. The hardest skill is consistency: producing a set that shares a style. Lock a detailed prompt skeleton, change only the subject, and reuse the same lighting and lens language every time.

1. 1Name the subject and what it's doing.
2. 2Add the setting and background.
3. 3Set the light and mood — this does most of the work.
4. 4State the lens/angle and aspect ratio; generate a few and refine one detail at a time.

Prompt formulas to copy