Start with the image’s purpose
Before choosing descriptive words, decide what the image needs to do. An illustration for an article may need empty space for a headline. A character portrait may need a clear face and a simple background. A product concept may need to show a particular shape. This purpose helps you choose which visual details belong in the prompt and which would only add noise.
Begin with a short statement of the subject, setting and intended use: “A ceramic watering can on a kitchen table, for a calm gardening newsletter illustration.” Then add directions about composition, lighting and color. These are more actionable than a long list of moods because they describe features you can inspect in the result. A prompt cannot guarantee that a model will follow every instruction, but specific visual priorities make revisions easier.
Use composition to guide the viewer’s eye
Composition describes how the subject and supporting elements are arranged within the frame. Specify the viewpoint, distance and placement when they matter. “Eye-level view,” “wide shot” and “close-up” suggest different relationships to the subject. “Watering can on the right third of the frame, clear space on the left” gives the model a layout goal that a request for “a beautiful watering can” does not.
Name the focal point and keep background instructions subordinate to it. For example: “The spout and curved handle are clearly visible; a softly detailed window and a few plants sit behind them.” If the output will be cropped for a banner or thumbnail, request a wide or square composition as appropriate, and leave breathing room around anything important. Even then, inspect the generated image at its final display size: a subject that looks clear in a large preview may disappear in a small thumbnail.
Describe what the light is doing
Lighting can communicate form, time of day and atmosphere. Instead of asking only for “dramatic lighting,” describe the light source and its effect: “Soft morning light enters from the left, casting gentle shadows across the table.” For stronger contrast, try “one narrow beam of light highlights the watering can while the background remains dim.” These descriptions give the model visible relationships to aim for.
Choose lighting that supports the image’s purpose. Soft, even light can make an object easier to read; directional light can emphasize texture and depth. Avoid stacking conflicting requests such as “deep shadows,” “flat lighting” and “every detail brightly lit” unless that tension is intentional. If a result feels too harsh or muddy, revise the lighting phrase first rather than replacing the whole prompt. That keeps the other useful parts of the composition intact.
Give color a role, not just a list
A color direction is most useful when it connects colors to parts of the scene. “Muted terracotta watering can, pale blue wall, warm cream tablecloth” is clearer than “terracotta, blue, cream.” You can also specify the overall relationship: a limited palette, low saturation, or one bright accent against restrained surroundings. These choices help establish hierarchy as well as mood.
Try to keep the palette compatible with the lighting. Warm window light may shift the appearance of a white tablecloth; a cool, overcast scene may make the same colors feel quieter. If exact colors are essential to a brand or product, treat generated images as exploratory rather than assuming a text prompt will reproduce those colors precisely. Check the result and adjust it in an editing workflow when accuracy matters.
Build, inspect and revise one decision at a time
A reusable prompt structure is: subject and setting; intended use; composition; lighting; color; essential constraints. Put it together like this: “Editorial illustration of a ceramic watering can on a kitchen table. Wide composition for a newsletter header, can on the right with open space on the left. Eye-level view; handle and spout clearly visible. Soft morning light from a window on the left, with gentle shadows. Muted terracotta can, pale blue wall and warm cream tablecloth. Uncluttered background.”
Generate an image, then compare it with the purpose you set at the start. Is there enough room for text? Is the focal point obvious? Does the light reveal the object, and does the palette feel coherent? Change one instruction that addresses the most important mismatch, then try again. If the layout works but the scene feels flat, revise the light rather than rewriting the subject and composition.
Finally, examine details rather than trusting the overall impression. Image models may render object parts, lettering or spatial relationships inconsistently. Use generated images as drafts, and check them before publication, especially when they depict a specific product or need to carry accurate text. A strong prompt improves your chances of a useful starting point; careful selection and editing make it fit the job.
