Gemini Nano Banana Image Prompts: The 2026 Guide

"Nano Banana" is the nickname that stuck to Google's Gemini image model, and Gemini nano banana image prompts are now one of the fastest-growing searches in AI image generation — for good reason. Unlike most image generators, Gemini is genuinely good at editing an existing photo while keeping the subject recognizable: same face, same outfit, same proportions, just a new scene, style, or action around them. That's a different skill than "generate a random image from text," and it needs a different kind of prompt.

This guide covers what Nano Banana actually does well, how to structure a prompt for it, and a set of ready-to-use examples you can copy straight into Gemini.

What "Nano Banana" actually is

Nano Banana isn't a separate app — it's the informal name the internet gave to Gemini's image generation and editing capability (the model card in Gemini and in Google AI Studio just says "Gemini 2.5 Flash Image" or similar). The name comes from an early leaked build, and it stuck because it's memorable. When people search "nano banana pro prompts," they mean prompts for Gemini's image model, full stop.

What makes it worth a dedicated prompting approach:

• Photo-preserving edits. You can upload a real photo and ask Gemini to change the scene, pose, or style around the subject while keeping their actual face and identity intact — something most text-to-image models can't do reliably.
• Instruction-following over vibes. Gemini responds well to explicit constraints ("NO distortion," "preserve exact proportions") rather than just mood words.
• Multi-element scenes. It handles prompts describing several specific elements at once — lighting, camera angle, props, action — better than very short, vague prompts.

How to structure a Gemini image prompt

A strong Nano Banana prompt usually has four parts, in this order:

1. Preservation instructions (if you're editing a photo) — tell it exactly what must stay the same: face, identity, outfit, proportions.
2. The scene or action — what's happening, where, with what objects.
3. Style and technical details — lighting, camera angle, aspect ratio, photorealistic vs. illustrated.
4. Explicit exclusions — what to avoid (distortion, extra fingers, cartoon look, face changes).

That fourth part matters more with Gemini than with most models — being explicit about what you don't want measurably cuts down on weird artifacts.

Ready-to-use Gemini Nano Banana prompts

1. Photo-preserving scene swap

Use the uploaded photo ONLY. Preserve the person 100% exactly — face,
identity, hair, glasses, outfit, colors and proportions. NO face change.

Place the same person in [new setting, e.g. "a cozy mountain cabin at
night, string lights, a mug of hot cocoa"]. Change ONLY the pose and
background — keep the subject's appearance identical.

Ultra-photorealistic, cinematic lighting, shallow depth of field,
realistic hands, believable scale, [aspect ratio, e.g. 9:16]. NO
distortion, cartoon look, or extra fingers.

2. Cinematic portrait generator

Create a cinematic portrait of [subject description]. Lighting:
[e.g. golden hour rim light / moody chiaroscuro / soft studio softbox].
Camera: [e.g. 85mm lens, shallow depth of field, slight low angle].
Mood: [e.g. quiet confidence, warmth, drama].
Style: photorealistic, high detail, realistic skin texture.
Aspect ratio: [e.g. 4:5]. Avoid overly smoothed or airbrushed skin.

3. Product or object showcase shot

Generate a clean product photo of [product]. Background: [e.g. soft
gradient studio background in a neutral tone]. Lighting: bright, even,
soft shadows. Angle: [e.g. three-quarter view]. Style: e-commerce
product photography, ultra-sharp focus on the product, subtle reflection
below it. No text, no watermark, no distortion of the product shape.

4. Stylized transformation (keeping identity)

Use the uploaded photo. Preserve the person's face and identity exactly.
Transform the rest into [style, e.g. "a hand-painted watercolor
illustration" / "a retro 1980s film poster" / "a claymation character"].
Keep proportions believable and the face clearly recognizable as the
same person. [aspect ratio]. No warping of facial features.

5. Text-only scene from scratch

A wide cinematic shot of [subject/scene], set in [location/time of day].
Lighting: [description]. Color palette: [e.g. warm amber and deep teal].
Include: [specific props or details you want visible]. Style:
photorealistic / editorial photography. Aspect ratio: 16:9. No text
overlays, no logos.

Each of these follows the same shape: preserve → describe → style → exclude. Swap the bracketed sections and you'll get a lot more consistent, usable output than a one-line prompt.

Common mistakes that ruin Nano Banana results

• Skipping the preservation clause on photo edits. If you upload a real photo and don't explicitly say "preserve face and identity exactly," Gemini treats the photo more like loose inspiration than a hard constraint, and the output can drift further from the original than you expected.
• Vague style words instead of concrete ones. "Make it look nice" gives the model nothing to grab onto. "Golden hour rim light, shallow depth of field, 85mm lens" gives it something it can actually execute.
• Forgetting the aspect ratio. Different platforms want different shapes — a square Instagram post, a 9:16 Story, a 16:9 thumbnail — and Gemini won't guess which one you need unless you say so.
• Piling on too many changes in one edit. Asking for a new background, a new outfit, and a new pose all at once tends to produce a muddier result than doing one change, checking it, then layering the next.

Tips for better Nano Banana results

• Be specific about "keep the same." Gemini is good at preservation, but only if you tell it precisely what to preserve — vague requests like "same person" leave room for it to drift.
• State the aspect ratio. If you don't specify one, you'll usually get a default square-ish crop that isn't ideal for social posts.
• One change at a time for edits. If you're editing a photo, asking for one clear change (new background, new outfit, new pose) gets more reliable results than asking for several at once.
• Describe the camera, not just the subject. Terms like "85mm lens," "low angle," or "shallow depth of field" push the output toward a more professional, intentional look.

If you'd rather skip writing these from scratch, PromptiFex has a fill-in-the-blank library of ready-made prompts built specifically for this kind of customization, including a dedicated collection of Gemini prompts you can adjust in a couple of clicks — see this Gemini face-swap style prompt and this photo-preserving coffee-cup scene prompt for two more real examples of the structure above in action.

More prompts to try

• Cinematic AI Image Prompt Generator — a reusable template for building your own cinematic image prompts.
• LinkedIn-ready Nano Banana Pro prompt — an example built specifically around Gemini's Nano Banana Pro image model.

Browsing the full prompt library is the fastest way to find more starting points across categories — most of the image prompts there include the same fill-in-the-blank fields shown above, so you don't have to rebuild the structure by hand each time.

Frequently asked questions

What is "Nano Banana" in Gemini? It's the popular nickname for Gemini's image generation and editing model (shown in-product as Gemini's image model, e.g. "2.5 Flash Image"). The name isn't official branding — it just caught on after an early preview — but it's now the term most people search for when looking for Gemini image prompts.

Do I need to upload a photo to use these prompts? No. Some of the prompts above (photo-preserving scene swaps, stylized transformations) are built around editing an uploaded photo, but others — like the cinematic portrait or product shot templates — work as pure text-to-image prompts with no upload needed.

Why does my Nano Banana output look distorted or change the person's face? Usually because the prompt didn't explicitly say what to preserve. Add clear preservation instructions ("preserve face, identity, and proportions exactly") and explicit exclusions ("no face change, no distortion") — Gemini follows those instructions much more reliably than it infers them on its own.