Course resource

Image Prompt Formula

A repeatable structure for getting the image you had in mind, and for getting it twice.

Tool names change; the formula does not. Check the Tool Radar for which generator currently leads.

The formula

[SUBJECT], [ACTION OR POSE], [SETTING], [LIGHTING], [CAMERA OR MEDIUM],
[STYLE], [COLOUR], [MOOD], [COMPOSITION]

You will not need all nine every time. Order matters: most generators weight earlier words more heavily, so the subject goes first.

Weak: a person working on a laptop

Using the formula:

A woman in her fifties, mid-sentence on a video call, home office with
bookshelves slightly out of focus behind her, warm afternoon light from a
window at camera left, shot on 50mm at f/2, documentary photography,
muted earth tones, calm and focused, medium shot with negative space on the right

The second gives you something usable. The first gives you stock-photo sludge.

Each slot, and what it controls

Slot Controls Examples
Subject What it is Be specific: age, build, clothing, expression
Action What is happening "mid-sentence", "reaching for", "having just"
Setting Where Include what is in the background and its focus
Lighting Mood, more than anything else golden hour, overcast, single hard source, backlit, neon
Camera Realism and depth 35mm, 85mm portrait, f/1.8, wide angle, macro, overhead
Medium What kind of image photograph, oil painting, pencil sketch, 3D render, risograph
Style The look documentary, editorial, Bauhaus, flat vector, watercolour
Colour Palette muted earth tones, high contrast black and white, two-tone
Mood Emotional register calm, tense, celebratory, lonely
Composition Framing rule of thirds, centred, negative space left, tight crop

Lighting is the highest-leverage word

If you change one thing to improve an image, change the lighting. It does more for the result than style keywords.

Getting consistency

The hard problem: the same character or style across multiple images.

For style: write a style block once and paste it into every prompt unchanged.

STYLE BLOCK (do not vary):
editorial photography, muted earth tones, soft diffused light,
shallow depth of field, 50mm, subtle film grain, no text

For a character: write a character block in the same way, and be more specific than feels necessary.

CHARACTER BLOCK (do not vary):
woman, mid-fifties, shoulder-length grey hair pulled back, round tortoiseshell
glasses, dark green cardigan over white shirt, small silver hoop earrings,
warm but serious expression

Vary only the action, setting and composition between images. Most inconsistency comes from unconsciously rewording the description each time.

Better still: where the tool supports reference images or character references, use those. Text alone will never be as consistent.

Negative prompts

Where supported, say what you do not want:

--no text, watermark, extra fingers, distorted hands, blurry, oversaturated,
lens flare, duplicate limbs

Text inside images has improved but is still the most common failure. If the image needs words, add them afterwards in a design tool.

Iterating properly

Change one variable at a time. If you change lighting, style and composition together, you learn nothing about which one helped.

A working loop:

  1. Get the subject and setting right. Ignore style.
  2. Fix the lighting.
  3. Add camera and medium.
  4. Add style and colour last.
  5. Adjust composition.

Keep the prompts that worked, with a note on what you changed.

Prompt What I changed Result

Before you publish

Back to dashboard