Better AI images usually come from better instructions, not longer instructions. The most reliable prompt formula is: subject + action + setting + composition + lighting + visual style + constraints. Start with the part that must be correct, describe the scene in a logical order, and add only details you can judge in the finished image. This guide shows how to turn a vague idea into a prompt that is easier to generate, compare and refine.
The AI image prompt formula
Prompt part | Question to answer | Example |
|---|---|---|
Subject | What is the main thing? | A ceramic coffee mug with a small blue mountain logo |
Action | What is happening? | Sitting beside an open notebook |
Setting | Where is it? | On a light oak desk in a quiet studio |
Composition | How is the frame arranged? | Three-quarter view, mug in the left third, negative space on the right |
Lighting | What creates shape and mood? | Soft window light from the left with gentle shadows |
Style | What should it look like? | Clean editorial product photography |
Constraints | What must be present or absent? | No hands, no extra cups, no text except the logo |
Put those parts together in one natural paragraph: “A ceramic coffee mug with a small blue mountain logo, sitting beside an open notebook on a light oak desk in a quiet studio. Three-quarter product view with the mug in the left third and negative space on the right. Soft window light from the left, gentle shadows, clean editorial photography. No hands, no extra cups, no text except the logo.”
Lead with the non-negotiable details
Image models do not treat every word as equally important. Put the subject, action and essential relationships first. If two people must be looking at each other, say that before describing the wallpaper. If a product label must face the camera, state that before adding atmosphere. This also makes failed generations easier to diagnose because the core request is visible at the start.
Avoid lists of adjectives that describe the same idea. “Beautiful, stunning, gorgeous, amazing” adds less control than “soft morning light, muted earth tones, shallow depth of field.” Concrete visual instructions are easier to evaluate than praise words.
Describe composition, not just content
A prompt can name every object and still produce a weak image if it says nothing about placement. Use framing terms that describe the final picture: close-up, waist-up portrait, wide establishing shot, overhead view, centered composition, rule of thirds, symmetrical layout, negative space, eye-level angle or low-angle view.
Goal | Useful composition language | Why it helps |
|---|---|---|
Website hero | Subject on the left, generous negative space on the right, wide 16:9 frame | Leaves room for a headline and button |
Product image | Centered object, three-quarter angle, seamless background, full product visible | Reduces cropping and distracting context |
Social post | Simple centered focal point, bold silhouette, square composition | Remains readable at small sizes |
Portrait | Head and shoulders, eye-level camera, eyes near the upper third | Creates a familiar professional framing |
Poster | Strong central subject, clear foreground and background, space for title at top | Builds a usable hierarchy |
Use lighting to control realism and mood
Lighting words often have more effect than broad style labels. Soft diffused light creates gentle transitions and is useful for portraits and products. Hard directional light creates stronger shadows and drama. Backlighting separates the subject from the background. Overcast daylight gives even exposure. Golden-hour light adds warmth, while cool fluorescent light can make a scene feel clinical.
Name one main light idea instead of combining several incompatible setups. “Soft north-facing window light” is clearer than asking for sunrise, neon, candlelight and studio strobes in the same scene.
Prompt examples: weak versus improved
Use case | Weak prompt | Improved prompt |
|---|---|---|
Portrait | Professional woman headshot | Head-and-shoulders studio portrait of a woman in a navy blazer, eye-level camera, neutral warm-gray background, soft key light at 45 degrees, natural skin texture, approachable expression, realistic corporate photography |
Product | Cool running shoe ad | Single white running shoe suspended above a pale blue surface, side profile, small splash of water beneath it, crisp rim light, high-speed commercial product photography, wide frame with empty space on the left, no text |
Interior | Modern cozy living room | Small modern living room with a low cream sofa, walnut coffee table and one large plant, late-afternoon window light, warm neutral palette, eye-level wide shot, realistic interior photography, uncluttered surfaces |
Illustration | Cute robot teaching kids | Friendly rounded robot showing three children a small solar-system model in a bright classroom, expressive faces, simple geometric shapes, soft pastel palette, children’s editorial illustration, no text |
Logo concept | Coffee shop logo | Simple one-color vector mark combining a coffee cup and crescent moon, bold rounded geometry, balanced negative space, black on white, no mockup, no gradients, no extra text |
Treat text as a separate production step
Image generators have improved at lettering, but exact wording, spelling and brand typography can still fail. For a poster, package or social graphic, generate the visual with deliberate empty space and add final text in a design tool. If text must appear in the generation, put the exact phrase in quotation marks, keep it short and request a simple front-facing layout.
Use constraints carefully
A short negative list can prevent common failures: no watermark, no extra fingers, no duplicate objects, no cropped product, no illegible text. Keep the list specific to the current image. A long catalogue of everything you do not want can compete with the positive description and make the prompt harder to interpret.
Build one variable at a time
When a result is close, revise only one category. First fix the subject and composition. Then adjust lighting. Then refine colour and style. If you rewrite every detail after each generation, you will not know which change improved or damaged the result.
Problem | Best next edit |
|---|---|
Wrong camera framing | Add shot size, angle and where the subject sits in the frame |
Scene feels generic | Add one specific setting detail and one lighting choice |
Too many objects | State the exact count and remove secondary props |
Subject looks artificial | Ask for natural skin texture, realistic proportions and restrained retouching |
Brand visual is inconsistent | Reuse the same palette, materials, camera angle and lighting phrase |
Important area is cropped | State “full object visible” and specify the aspect ratio |
Match the prompt to the output
A thumbnail needs one clear subject and strong contrast. A website hero needs negative space. A product listing needs a clean background and accurate shape. A logo concept needs flat geometry rather than a photographed mockup. Decide where the image will be used before describing its style.
If you are still choosing a model, compare the strengths and workflows in our AI image generator ranking. For marks and identity concepts, see the AI logo generator comparison.
A reusable prompt template
[Main subject] [action or pose] in/on [setting]. [Shot size and camera angle], with [subject placement and negative space]. [Main lighting choice], [colour palette], [visual medium or photographic style]. [Important materials or details]. [Aspect ratio or output purpose]. No [specific unwanted elements].
The template is a checklist, not a requirement to fill every slot. Omit details that do not matter. A controlled prompt gives the model enough direction to satisfy the brief while leaving room for visually coherent choices.
Frequently asked questions
How long should an AI image prompt be?
Long enough to define the subject, composition and visual treatment. Many strong prompts fit in 40 to 100 words. More words help only when they add usable visual information.
Should camera settings be included?
Use them when you understand the effect you want. “Shallow depth of field” is often clearer than a random aperture value. Physical camera terms can guide the look, but they do not guarantee physically exact rendering.
What should I change first when an image is wrong?
Fix the subject and composition first. Style cannot rescue the wrong object, pose or crop. Change one variable, compare the result, and keep the instructions that worked.
Final checklist
Before generating, confirm that the prompt names one clear subject, explains the required action or relationship, defines the framing, chooses one main lighting idea, states the intended visual style and includes only relevant constraints. The best prompt is not the most elaborate. It is the one that makes success easy to recognize.