Describing an Image So You Get It
A repeatable structure for image prompts, why adjective piles fail, and how to steer a result you are almost happy with.
The instinct is to write 'beautiful stunning professional high quality photo of a coffee shop'. Four of those seven words carry no information — every image in training was captioned by someone who thought it was fine, so 'stunning' does not point anywhere in particular.
Useful prompts describe what a photograph of the thing would contain.
The six slots
You do not need all six every time, but knowing which one is missing is how you fix a disappointing result.
- Subject — what it is, specifically. 'A ceramic pour-over coffee dripper', not 'coffee equipment'
- Action or state — what is happening. Steam rising, hands mid-pour, still on a shelf
- Setting — where, and what else is in frame
- Lighting — the single highest-impact word group. Soft morning window light, harsh midday sun, warm tungsten, overcast
- Composition — close-up, wide shot, overhead flat lay, shallow depth of field, negative space on the left
- Medium and style — photograph, watercolour, flat vector illustration, 35mm film, 3D render
The same idea, described two ways
Weak:
beautiful professional photo of a coffee shop, high quality, 4k
Strong:
Overhead flat lay of a ceramic pour-over dripper on a pale
oak table, steam rising, a linen cloth and scattered coffee
beans to one side. Soft diffused morning light from the left,
gentle shadows. Shallow depth of field. Muted warm palette,
generous empty space in the upper right for text.
Photograph, natural colour.Negative space is the most useful thing you can ask for
If the image will carry a headline, a logo or a button, ask for the empty area explicitly. 'Generous empty space in the upper right' gets you something you can actually design on top of, instead of a beautiful image with the subject exactly where your text needs to go.
Common failures and the fix
- Too busy → name fewer objects and ask for a plain or blurred background
- Looks like generic stock → add specificity: a material, a time of day, a small imperfection
- Wrong mood → change the lighting words before anything else; lighting carries mood more than colour does
- Faces look uncanny → shoot wider, from behind, or crop out the face. Hands and faces are the hardest surfaces
- Style drifts between images → move the style words to the front and keep them identical across generations
- Too saturated and glossy → ask for muted, natural colour, film grain, or overcast light
Reference images beat words
Most tools now accept an image as input for style or composition. When you have a reference — a previous image of yours, a mood board frame, a rough sketch — use it. A reference communicates a look in one step that would take you six rounds of adjectives to approximate, and it is the only reliable way to keep a series consistent.
What to take from this chapter
- Describe subject, action, setting, lighting, composition and medium — skip empty quality words
- Lighting words change mood more than any other part of the prompt
- Ask explicitly for negative space when text or a logo will go on top
- Refine by changing one element; rewriting gives you a new image, not a better one
- Use a reference image whenever you have one — it beats adjectives for style and consistency
Try it
Take a prompt that disappointed you and rewrite it using the six slots, adding nothing but specifics. Generate both. The difference between them is the whole lesson.