Every strong image prompt is really six decisions in a row. Make them on purpose, in this order, and you can build any scene deliberately instead of hoping. Miss one and the model fills the gap with its blandest average — which is exactly why generic AI images look generic.
1 · Subject — specific beats pretty
Name the thing exactly. Not "a candle" but "a hand-poured soy candle in an amber glass jar with a cream label." The model can only render what you name; every unnamed detail becomes a guess.
2 · Setting — where it lives
Put it somewhere real: "on a linen-draped table with a few eucalyptus stems, in a bright morning kitchen." Setting is what turns a product shot into a lifestyle scene.
3 · Style — the visual world
Tell it the genre: "editorial product photography," "soft film photo," "minimal flat-lay." This one word does more than any other to set the price-bracket the image reads as.
4 · Light — 50% of "expensive-looking"
Underrated and decisive: "soft window light from the left, gentle shadows." Cheap images almost always have flat, direct, phone-flash light. Name the light and you're halfway to the magazine look.
5 · Lens & camera cues — the photographer's grammar
Borrow the vocabulary of real photography: "shot on 85mm, shallow depth of field, subject in focus, background softly blurred." These cues tell the model to imitate genuine cameras, not clip art.
6 · Format — aspect ratio by destination
Decide the shape before you generate, because it's set by where the image is going: Pinterest 2:3, Instagram feed 4:5, Instagram story 9:16, blog header 16:9, square 1:1. Generating in the wrong ratio means awkward cropping later.
Same candle, six prompts — watch the diff
Here's the whole method in one worked example: one product, six prompts, each changing only a slot or two. Read the italic line under each to see exactly what moved. Copy any of them and swap in your own product.
The baseline. Subject and setting are vague, style is generic, light is flat. This is what "a candle" gets you — serviceable, forgettable.
Diff: Subject got exact (amber jar, cream label); setting got a surface (marble); style named ("product photography"). Same light, but already looks like a real listing.
Diff: Setting became a scene (linen, eucalyptus, morning kitchen); light got named and directional (window, from the left). This is the jump from "product" to "brand."
Diff: Style upgraded to "editorial"; lens cues added (85mm, shallow depth of field). The background softens, the candle pops — the photographer's grammar doing its job.
Diff: Setting stripped back to negative space + plaster; style shifted to "quiet-luxury minimalist still life"; light moved overhead and diffused. Same candle, now reads high-end and calm.
Diff: Setting got art-directed (brass tray, styling); style named a use ("for a luxury home brand"); light became dramatic; texture cue added (film grain). This is the magazine version — and it's the same six slots, just fully specified.
Nothing changed between #01 and #06 except specificity in the six slots. You didn't get more artistic — you got more precise. That's the whole skill, and you now have it.
Negative prompts: say what you don't want
Sometimes the fix is subtraction. Many tools accept a "negative" clause (or you just phrase it in-line) to rule out the junk that keeps creeping in. Common ones for brand work:
- Avoid: text, watermark, logo, extra objects, clutter, harsh flash, plastic-looking, oversaturated, distorted, low-resolution.
- For faceless-brand shots: "no faces, hands only, natural skin."
- For clean flat-lays: "no props overlapping the product, even margins."
Iteration etiquette: change ONE slot at a time
This is the single habit that separates people who "can't get AI to do what they want" from people who reliably can. When an image is almost right, resist the urge to rewrite the whole prompt. Change one slot, regenerate, compare. Warmer? Keep it and change the next. You're tuning a dial, not rolling dice — and if your tool lets you hold the seed (Lesson 1), the rest of the image stays put while your one change takes effect.
Take your own product (or the "before" candle from Lesson 1) and write it up the ladder: start at prompt #02 above with your subject swapped in, generate, then add setting, then name the light, then add a lens cue — one slot per generation. Watch it climb. Save the version you'd actually post. You just built an image on purpose.