Image prompting basics: say what you see, in the right order

A reusable prompt template with slots for subject, light, camera and mood, and five consistent variations of it.

Best for: Anyone starting with image models who gets random results and wants a method instead of luck.

• Access to at least one image model.
• A clear purpose for the image (product, cover, ad, background).
• A place to keep prompts and results together so you can compare.

1. 1

   Start with the subject and the action

   State what is in the picture and what it is doing in plain words. A short, concrete noun phrase beats a long list of adjectives.

2. 2

   Describe the light

   Light is the biggest lever. Say where it comes from, how hard or soft it is and what colour. “Soft window light from the left” and “hard rim light in copper” give very different pictures.

   ✦Tip. If a result looks flat, fix the light before changing anything else.

3. 3

   Describe the camera

   Distance, angle and lens: close-up, eye level, wide angle, shallow depth of field. You are telling the model where to stand.

4. 4

   Add mood and palette

   Two or three mood words and a limited palette keep a set consistent. Name colours instead of saying “colourful”.

5. 5

   Leave room and say what to exclude

   If you will add text or a product later, ask for empty space and where it should be. Add no text, no logos and no watermark, since models draw gibberish letters.

6. 6

   Change one thing at a time

   When a result is wrong, edit one part of the prompt and generate again. Keep a note of what changed, and you will learn what each part does in your tool.

7. 7

   Turn the winner into a template

   Replace the subject with a bracketed slot and keep the rest. You now have a recipe that produces a consistent set instead of one lucky image.

Replace anything in [square brackets]. Prompts are starting points: results vary between tools and versions, so review every output.

Prompts are written in English because image, video and most text models follow English instructions best. To get answers in your language, add “Reply in English.” to a prompt for a text assistant.

[Subject and action], [setting], [lighting: direction, quality, colour], [camera: lens, distance, angle], [mood or style in plain words], [composition and empty space], [colour palette], no text, no logos, no watermark.

Here is an image prompt that gave a poor result: "[prompt]". The problem I see: [problem]. Rewrite it three ways: more specific about the subject, more specific about light and camera, and simplified to the 3 most important ideas. Explain what each change should do.

Turn this loose idea into a reusable prompt template with bracketed slots: "[idea]". Then give me 5 filled-in variations that keep the style consistent but change the subject, so the whole set looks like one collection.

• ✕Writing a long list of adjectives with no clear subject.
• ✕Ignoring light and then wondering why the image feels flat.
• ✕Changing five things at once, so nothing is learnt.
• ✕Trusting generated text inside an image.