Text to image AI: how to write prompts

Start with the subject: “a ceramic mug on a wooden table.” Add style: “product photo, soft studio light.” Add constraints: “white background, 4:5 aspect.”

Put the most important words first. Models weight early tokens heavily.

Name a medium if you care: “oil painting,” “flat vector,” “35mm film photo.”

For portraits, specify age range, expression, and wardrobe instead of vague words like “beautiful.”

Do not stack ten adjectives. Pick three that matter.

Negative prompts (what to exclude) help on some models. In UltraPix, check if the screen exposes negative text for your selected model.

Save prompts that worked. Change one variable per run: lighting, then lens, then color palette.

Use image-to-image when you have a reference photo instead of describing every detail in text.

No reference photo yet? Start from text-to-image templates. Already have the shot? Switch to image-to-image patterns and describe only the delta.

Name direction and quality: “soft window light from camera left,” “hard noon sun, short shadows,” “overcast open shade.”

Color temperature matters for product shots: “5500K neutral studio” vs “tungsten practicals, warm amber spill.”

For portraits, specify catchlights (“large softbox reflection in eyes”) to avoid flat mugshot lighting.

State crop intent: “4:5 Instagram portrait,” “16:9 YouTube thumbnail, subject right third,” “1:1 marketplace main image.”

Negative space instructions help ads: “empty top third for headline,” “subject lower half, sky gradient above.”

Model-specific guides such as Nano Banana Pro and Nano Banana Pro prompting show how reasoning models interpret layout instructions.

Lead with material and finish: “matte black aluminum bottle, brushed cap, condensation beads.”

Name marketplace rules when relevant: “Amazon main image, pure white seamless, product fills 85% frame.”

See AI product photography and product prompt examples for catalog-ready language.

When the UI exposes negative text, use it for recurring failures, extra fingers, watermark patterns, cluttered backgrounds, not for long wish lists.

Reference uploads beat adjectives for brand colors and silhouette. One photo of the product angle you need saves fifty words of description.

For edits, start prompts with lock phrases: “keep face and pose, change only jacket color to navy.”

If results drift, reduce prompt length and add one reference image instead of five more adjectives.

Subject first, style second, lighting third, aspect last.

One change per iteration when learning a new model.

Quote on-image text verbatim.

Save winners in the app history, not only in chat threads.

Use img2img when camera angle is right but mood is wrong.

Read one model-focused blog monthly for fresh vocabulary on Android.