Nano Banana Prompts
Below are prompts tested on Nano Banana (Google Gemini) that produce strong, predictable output. Each one includes the model setting and aspect ratio that work best. Click "Try" to run it in the Noaroo playground.
Examples
Prompts to try
Photoreal headshot
studio headshot of a 35-year-old woman, slight smile, navy turtleneck, soft beauty dish lighting, 85mm f/1.4, neutral grey backdrop
Product hero shot
matte black mechanical keyboard on a walnut desk, single overhead key light, deep shadow, top-down 3/4 angle, photoreal
Lifestyle scene
young couple walking dog along brooklyn brownstone street, autumn leaves, golden hour, kodak portra film grain
Anime portrait
anime portrait of a swordswoman in a windswept field, studio ghibli color palette, watercolor backgrounds, expressive eyes
Architectural render
modern minimalist beach house at dusk, floor-to-ceiling glass, warm interior glow, photoreal architectural visualization, wide angle
Pixel art game scene
16-bit pixel art of a cozy alchemist shop interior, warm candlelight, shelves of potions, isometric view
Cinematic action
samurai mid-strike on a rain-soaked rooftop, neon signs reflecting in puddles, motion blur, cinematic 35mm
Stock-style food
overhead flat lay of brunch table: eggs benedict, fresh berries, latte, linen napkin, natural window light
How to read a Nano Banana prompt
Three things matter most for this model:
- Subject first, then action / pose, then setting, then lens / lighting, then style.
- Concrete nouns over adjectives. "Brass tea kettle" beats "shiny metallic kettle".
- Lens / lighting language is worth 30% of the result quality. Always include focal length or light direction for photoreal output.
When to use sh-flash vs sh-pro
- flash: ideation, multiple variations, anything 512px or 1K square.
- pro: portfolio shots, faces close-up, anything you'll publish.
Editing prompts
For edit mode, upload the image and describe only the change: "change shirt color to navy", "remove the person on the left", "add neon signage to the building". Don't re-describe the whole scene.
FAQ
Why don't my prompts give the same result twice?
Image generation is non-deterministic. Same prompt produces different outputs each run. To stay close to a reference, use edit mode with the previous output as input.
Should I use negative prompts?
Nano Banana doesn't use negative prompts the way Stable Diffusion does. Just describe what you want; the model handles avoidance implicitly.
How long should a prompt be?
60-120 words is the sweet spot for photoreal scenes. Going past 200 starts diluting attention.