How to Write AI Image Generator Prompts (With Examples)

A good AI image generator prompt names four things: the subject, the setting, the light, and the style. Get those down in one or two plain sentences and most generators will return something close to what you pictured. Skip them and you get the model's default idea of your subject, which is rarely the one you wanted.

This guide walks through the formula, then shows it applied to portraits, product shots, logos, and photo edits. Every example below was written for a text-to-image model like Nano Banana 2 or Nano Banana Pro, and you can paste any of them straight into the generator to see the result.

The four-part formula for AI image generator prompts

Structure beats length. A prompt of 20 to 60 words with a clear order — subject, setting, light, style — usually outperforms a rambling paragraph, because the model weighs early words more heavily than late ones.

1. Subject. Who or what, plus one or two details that matter: "a barista in her fifties", "a matte black ceramic mug". Vague subjects produce vague images.
2. Setting. Where the subject sits: "in a sunlit Lisbon café", "on a concrete studio table". The setting also implies a camera distance, so say "close-up" or "wide shot" when you care.
3. Light. The single most underused lever. "Soft window light", "overcast afternoon", "hard flash at night" — light decides whether the same subject looks like a stock photo or a magazine spread.
4. Style. The look of the output: "35mm film photo", "watercolor illustration", "flat vector logo". Naming a medium keeps the model from guessing.

Mood words ("calm", "dramatic", "nostalgic") can help, but they work only when the concrete parts are already in place.

Example prompts for portraits and people

Portraits are where prompt order shows the most. Compare these two:

"A woman on a street."

"A close-up portrait of a woman in her sixties, laughing, golden-hour side light, shallow depth of field, 85mm lens look."

The first gives the model every decision to make. The second makes the decisions for it. A few more you can copy:

• "Full-body shot of a skateboarder mid-trick, skatepark at dusk, motion blur in the background, shot on 35mm film."
• "A grandfather and a small dog sharing a bench, soft overcast light, muted colors, documentary photography style."
• "Studio portrait of a jazz trumpeter, single hard key light from the left, deep shadows, black background."

One caution that applies to every generator: describing real, private people in invented situations causes more problems than it solves. Keep likeness work to photos you have the rights to — image-to-image editing with your own reference is the right tool there, not a text prompt.

Example prompts for products and e-commerce

Product images sell objects, so precision about materials and surfaces pays off directly:

• "A pair of white running shoes on a light-gray seamless background, soft studio light from above, crisp product photography."
• "A glass perfume bottle on wet stone, backlit, water droplets, luxury advertising photography."
• "Handmade ceramic bowl, top-down view on a linen cloth, window light from the right, warm and slightly grainy."

For e-commerce use, add the background you actually need ("seamless white", "in a kitchen scene") rather than removing backgrounds later. The showcases page has more product-design examples, each linked to the prompt that made it.

Example prompts for logos and graphics

Logo work needs a style constraint, or the model drifts toward glossy 3D renders. Two that stay usable:

• "Minimal single-line logo of a coffee bean and steam, black on off-white, flat vector, circular badge."
• "Isometric illustration of a small city block, soft pastel palette, clean edges, no text."

Adding "no text" is deliberate — image models still misspell words often enough that any lettering is better added in a design tool afterward.

Editing an existing photo instead of generating one

When you already have the right photo, editing beats regenerating. Upload the image and describe the change the same way: "make the background a soft warm gradient, keep the person and the lighting unchanged". The edit keeps the original scene intact, which matters for things like trying a new hairstyle on your own photo — you want your face back, with different hair.

Two habits make edits more reliable. Name what should stay the same ("keep the pose and the outfit"), and make one change per generation rather than stacking five. Iterating is cheap: an image costs single-digit credits on most models here, so running three focused passes beats one overloaded prompt.

Quick checklist before you generate

• Subject, setting, light, style — all four named, subject first
• 20 to 60 words, no filler adjectives stacked five deep
• Camera language only when you want a photographic look
• "No text" on anything with lettering
• For edits: one change per pass, and say what stays fixed

FAQ

How long should an AI image prompt be? Between 20 and 60 words is the sweet spot for most models. Long enough to cover subject, setting, light, and style; short enough that every word carries weight. If a prompt needs a paragraph, it usually needs to be split into several images instead.

Do I need to write prompts in English? Most image models, including the Nano Banana family, understand English prompts best. Write the prompt in English even if the rest of your work is in another language — you can always ask for a subject "with a Chinese street market setting" in English words.

Why does the same prompt give different results each time? Generative models sample from a probability distribution, so each run draws a slightly different interpretation. That is a feature for exploring variations and a reason to pin down light and style when you want consistency. Adding "same composition" and reusing a seed, where the tool offers one, tightens the spread.

Can I use the generated images commercially? Yes, images generated here can be used commercially. Make sure you hold the rights to any reference photos you upload, and check the terms of service for the specifics.

Conclusion

Writing a better prompt is mostly writing a more specific one: subject, setting, light, style, in that order, in one or two sentences. The examples above cover the common cases, and the fastest way to internalize the pattern is to take one and change a single element at a time. Open the AI image generator, paste a prompt, and start from there.