Open Soulris

Writing Effective Image Prompts — Subject, Style, Composition, Lighting

October 11, 2026 · Soulris Team · 3 min read · images

Why prompt wording matters

An image model only has your words to work with — there's no back-and-forth clarification before it generates, the way there can be in a text chat. A vague prompt gets a generic result; a specific one gets closer to what you actually pictured. The good news is that "specific" doesn't mean "long" — it means covering the right handful of details.

Four things consistently make the biggest difference: subject, style, composition, and lighting. You don't need all four in every prompt, but knowing what each one controls helps you fix a result that's close but not quite right.

Subject: say exactly what and who

Start with the core subject, and be concrete rather than abstract. "A dog" is a start; "a golden retriever puppy sitting in grass" tells the model the breed, age, pose, and setting in one phrase.

If the subject has details that matter to you — color, clothing, expression, specific objects nearby — say so directly. The model can't infer details you haven't mentioned; it fills gaps with whatever's statistically common for that kind of image.

Style: tell it what kind of image you want

Without a style cue, the model defaults to something generic. Naming a style narrows that down a lot:

Style words have an outsized effect on the overall look, so this is often the single most useful addition to a prompt that's coming back too generic or "AI-looking."

Composition: describe the frame, not just the subject

Composition is about how the subject sits in the image — this is where a lot of prompts fall short because it's easy to describe what's in the image and forget how it's arranged.

Useful composition cues:

Lighting: it changes the mood more than almost anything else

Lighting is the detail most people forget, and it's often the difference between a flat result and one that feels intentional.

The same subject and composition, lit differently, can read as a completely different image — a product photo under studio lighting versus the same product under warm evening light tells two different stories.

Putting it together

A prompt that stacks all four reads naturally, not like a checklist:

"A close-up photorealistic photo of a steaming cup of coffee on a wooden table, soft morning light from a window, shallow depth of field, blurred background."

Subject (coffee cup), style (photorealistic photo), composition (close-up, shallow depth of field, blurred background), lighting (soft morning light) — all in one readable sentence.

When a result isn't working

If what comes back is close but off, change one element at a time rather than rewriting the whole prompt — swap just the lighting, or just the style word, and regenerate. That tells you which part of the prompt was actually driving the thing you didn't like. See regenerating or tweaking an image for how to iterate on a result you already have rather than starting from scratch each time.

Match your style word to the right quality tier

A photorealistic prompt tends to show the difference between quality tiers more visibly than a simple flat illustration does — see choosing an image quality tier and what it costs if you're deciding how much quality a particular style of image actually needs.

Try it yourself

Pick one image idea and generate it three times — once with no style word, once with a style word, once with style and lighting both — and compare. It's the fastest way to see how much these details actually move the result. Head to Image Studio or start a chat and ask the AI to generate an image to try it.