How to prompt Nano Banana

Generate a high-end commercial beauty shot of a sleek nude-colored face moisturizer jar on wet slate, soft radiant lighting, dew drops catching the light, macro detail on the jar's surface. Render the text "GLOW DIFFERENT" in a bold white sans-serif centered at the top, and "New formula" in a smaller light script at the bottom-right.

Quoted text + a described font + an explicit position is the reliable recipe for typography. Two short phrases render cleanly; a paragraph of copy will not — keep on-image text under 10-15 words.

Edit this photo: remove the parked car on the right side of the street and fill the space with the same wet cobblestone and storefront reflections. Keep everything else pixel-identical — the people, the lighting, the neon signs, the rain.

This is semantic masking: you define the region with words instead of a brush. The "keep everything else" clause is not decoration — being explicit about what must not change is what keeps the rest of the frame stable.

Using the attached reference images of Maya — the woman with the copper bob and round glasses — generate her mid-laugh at a farmers market stall, holding a paper bag of oranges, candid street-photography style, morning light, 50mm lens. Same face, same glasses, same hair as the references.

Name the character and restate the identity anchors ("same face, same glasses") — cheap insurance against drift. Reuse the identical reference set for every image in the series and the character holds across all of them.

Edit this product banner: find the headline "SUMMER SALE" at the top-center and replace it with exactly "FINAL DROP" in the same font, weight, color and size. Do not alter the product, background, layout or any other element — text change only.

Text replacement inside an existing graphic is a Nano Banana specialty. Spell the replacement exactly, point at the position, and demand the original style — the edit preserves the source image's dimensions automatically.

A photorealistic portrait of a fisherman in his 60s on a foggy dock at dawn, weathered skin with visible pores, salt-and-pepper stubble, wool sweater with loose threads, 85mm portrait lens at f/1.8, soft directional light from the left, muted cold palette. The image must be indistinguishable from a real photograph.

Asking for imperfections — pores, stubble, loose threads — is what kills the plastic AI look. Lens and aperture language ("85mm at f/1.8") gets you real depth-of-field behavior instead of a generic blur.

Using the attached napkin sketch for the structure and the attached fabric swatch for the upholstery texture, transform this into a high-fidelity product render of an armchair, placed in a sun-drenched minimalist living room, soft window light, editorial interior-photography style.

Each reference gets a stated job — structure from one, texture from the other. That "relationship instruction" is what makes multi-image composition predictable; without it the model averages the inputs.

Edit this photo: keep the sneaker exactly as it is — same shape, colors, laces, stitching, logo placement — and replace the background with a seamless burnt-orange studio backdrop, soft shadow under the sole, subtle rim light on the heel.

Product fidelity comes from the keep-list: enumerate what defines the product and the edit will not redraw it. The new background is described like a fresh generation — surface, shadow, light.

Transform this photograph of the Amalfi coast into a loose watercolor travel-journal illustration: soft bleeding washes, visible paper grain, hand-drawn ink outlines on the buildings, keep the composition and the boats exactly where they are.

Style transfer is an edit with a medium change: name the target medium and its physical tells (bleeding washes, paper grain, ink outlines). Pinning the composition keeps the recognizable geography intact.

Create a clean step-by-step infographic titled "COLD BREW AT HOME" with four numbered steps, each with a small flat icon and one short caption: "1. Grind coarse", "2. Steep 18h", "3. Strain slow", "4. Serve over ice". Warm cream background, single accent color in burnt orange, modern editorial layout, bold sans-serif headline.

Layouts with multiple labeled elements are where the reasoning tier earns its keep — Nano Banana Pro and 2 plan the arrangement before drawing. Keep every caption to a few words and quote each one.

Search for today's weather in London. Generate a miniature diorama of the London skyline inside a glass teacup that reflects the current conditions — if it is raining, tiny rain clouds and wet reflections; if sunny, warm golden light. Photorealistic macro shot, shallow depth of field.

With the Web Search toggle on, Nano Banana Pro and 2 can pull live facts into the image. Structure it as search → interpret → visualize so the model knows the data changes the scene, not just decorates it.

A lone hiker crossing a ridgeline at dusk, vast layered mountains behind, tiny warm headlamp glow, cinematic color grading with muted teal shadows and amber highlights, shot as if on 1980s color film with slight grain, wide-angle lens, aerial three-quarter view.

Film-stock and grading language ("muted teal shadows", "1980s color film") steers the palette more reliably than mood adjectives. 21:9 turns it into a ready-made site hero or channel banner.