How to write AI image prompts

A useful AI image prompt is a compact creative brief. It tells the model what the image must communicate, how the scene should be organized, and which visual constraints matter. It does not need to be a wall of adjectives. In most projects, a clear hierarchy produces a better starting point than a long list of loosely related styles.

This guide uses a repeatable structure:

1. Purpose and subject — what the image is for and what must appear.
2. Scene and composition — where elements sit and how the viewer reads them.
3. Light, color, and material — the visual decisions that create mood and physical character.
4. Format and constraints — aspect ratio, text, background, exclusions, and details that must remain stable.

You can study these parts in the AI image prompt library, then test a revised brief in the Wenimg image generator.

1. Start with the job the image must do

Compare “a premium bottle on a table” with “a wide ecommerce hero image for a premium skin-care bottle, product centered in the right third, quiet background and open space on the left for a headline.” The second prompt gives the model a communication goal and a layout decision. Those details matter more than adding several photography buzzwords.

Write one sentence that answers three questions:

• Where will the image be used?
• What is the main subject?
• What should the viewer notice first?

For an editorial portrait, the face and expression may be primary. For a product listing, accurate shape and readable packaging may matter more. For a poster, the visual hierarchy and text-safe area can be part of the requirement.

2. Describe composition with visible relationships

Composition instructions work best when they describe relationships that can be seen: close-up or wide shot, eye level or overhead, centered or asymmetrical, foreground and background, open space, depth, and crop.

Example:

Overhead food photograph of a ceramic bowl of tomato noodles, bowl placed slightly below center, chopsticks entering from the upper right, two small ingredient dishes in the background, generous negative space around the plate, square composition.

This is easier to evaluate than “beautiful cinematic food photo.” If the first result is too busy, remove a background object. If the crop is wrong, change only the framing instruction. Controlled revisions reveal which phrase actually changes the image.

3. Use lighting, color, and materials to resolve ambiguity

Lighting establishes direction, contrast, and time of day. Color creates hierarchy and brand fit. Material words help the model distinguish polished metal from brushed aluminum, translucent glass from opaque plastic, or soft cotton from glossy satin.

Prefer concrete instructions such as:

• soft window light from camera left with a gentle shadow;
• warm ivory background with one cobalt accent;
• matte recycled paper with visible fibers;
• clear glass edges with realistic refraction;
• restrained contrast suitable for an ecommerce listing.

Avoid stacking conflicting directions such as soft flat light, dramatic hard shadows, pastel color, neon color, minimal layout, and maximal detail in the same brief. Decide which two or three characteristics serve the purpose.

4. State exact constraints once

Put non-negotiable requirements near the end of the prompt. Examples include the output ratio, exact text that must appear, background color, number of objects, or a feature that must not change when editing a reference image.

When using a reference, explain its role: “keep the bottle shape and label placement from the reference, but replace the environment with a bright bathroom shelf.” This is clearer than asking the model to “use this image” without saying what should remain stable.

5. Revise from the result, not from memory

After generation, compare the output with the original job. Name the largest mismatch and update one or two instructions. A useful review order is:

1. Is the correct subject present?
2. Does the composition support the intended use?
3. Are identity, product, or text constraints respected?
4. Do lighting, color, and material support the mood?
5. Which single change would improve the next attempt most?

Keep successful structure and replace only project-specific details. Over time, you will build prompt templates for product heroes, portraits, posters, illustrations, and other recurring formats.

A complete prompt example

Ecommerce hero image for a compact silver espresso machine. Three-quarter front view, machine placed in the right third, one ceramic cup beneath the spout, pale travertine counter and uncluttered warm-gray wall. Soft morning window light from the left, realistic brushed-metal texture, gentle contact shadow, premium but approachable mood. Wide 16:9 layout with clean negative space on the left for a two-line headline. No extra appliances, no floating objects, no text rendered in the image.

Use this as a structure, not a fixed formula. Replace the purpose, subject, spatial relationships, surface choices, and constraints with the decisions your own image needs.

Model behavior and supported controls change over time. For implementation-specific details, consult the OpenAI image generation guide, Google Gemini image generation documentation, and the current controls shown in Wenimg.