Nano Banana 2 Prompt Library
Copy-paste prompt templates for Google's Nano Banana 2, organized by the jobs creators actually run.
Good Nano Banana 2 prompts are the difference between a model that feels magic and one that feels random. Nano Banana 2 is Google's image model built on Gemini 3.1 Flash, and it is unusually literal: it follows precise instructions, accepts up to fourteen reference images, outputs anywhere from 0.5K to 4K, and can even ground a generation in real-world references it looks up. That precision rewards prompts that are specific and punishes ones that are vague. This library gives you more than thirty copy-paste Nano Banana 2 prompts, grouped by the jobs you actually run, so you can paste, swap in your subject, and ship.
Every template below is written to be adapted. Replace the bracketed parts with your own subject, keep the structure, and you will get consistent results. If you want to run them as you read, you can paste any of these straight into Nano Banana 2 and iterate in the browser.
What makes a good Nano Banana 2 prompt
A reliable Nano Banana 2 prompt has four parts in order: the subject, the instruction, the style, and the constraints. The subject is what you are editing or generating. The instruction is the single change you want. The style covers light, lens, and mood. The constraints lock down aspect ratio, what to keep, and what to avoid. Keep those four in that order and the model rarely misreads you.
Nano Banana 2 also responds well to plain, declarative language. You do not need to stack ten adjectives; you need one clear instruction and a couple of concrete details. Because the model is strong at following edits on an existing image, phrases like keep everything else unchanged are doing real work — they tell it where not to wander. The model is fast enough that you can afford to test a phrasing, read the result, and tighten it, rather than trying to write one perfect mega-prompt up front.
Nano Banana 2 vs Pro: which to prompt
The common question is Nano Banana 2 vs Pro, and the honest split is this. Nano Banana 2 delivers most of the Pro variant's quality at roughly two to three times the speed, which makes it the default for drafting, iteration, and the large majority of edits. Pro earns its slower render when the output is brand-critical, text-heavy, or layout-sensitive and needs to be production-final in one pass. A practical rule: draft and explore with Nano Banana 2, and reserve Nano Banana Pro prompts for the final, polished version when the stakes are high. For a deeper breakdown, our full Nano Banana Pro guide compares them shot by shot.
Nano Banana 2 prompts for photo editing
Nano Banana prompts for photo editing are the most common reason people reach for this model, because it edits a real photo without redrawing the whole thing. The nano banana prompts examples below cover the everyday retouching jobs. Upload your photo first, then paste and adjust the bracketed details to fit your shot.
- Background replace:
Using the attached photo, replace the background with a softly blurred city street at dusk, keep the subject, their pose, and the lighting on their face unchanged, match the new background color temperature to the existing light - Relight:
Relight this portrait with warm golden-hour light coming from the left, add a gentle rim light on the hair, keep the subject and composition identical, do not change the face - Object removal:
Remove the trash can and the street sign from this image, reconstruct the wall and pavement behind them naturally, keep everything else unchanged - Object add:
Add a steaming coffee cup on the table in front of the subject, match the perspective and the room lighting, keep the rest of the photo unchanged - Retouch:
Lightly retouch this portrait: even the skin tone, reduce shine on the forehead, keep natural skin texture and pores, do not smooth into plastic, leave the eyes and hair untouched - Color grade:
Apply a muted cinematic color grade to this photo, teal shadows and warm highlights, slightly lifted blacks, keep detail in the sky, preserve the subject's natural skin tone
For edits where the change has to land exactly on a small region, the model's precision editing controls help you keep the rest of the frame frozen while one element changes. That control is what makes editing prompts feel surgical rather than like a full re-roll of the image.
Nano Banana 2 prompts for character consistency
Character consistency is where reference images earn their keep. Because Nano Banana 2 accepts up to fourteen references, you can hand it several angles of the same person and ask for a new scene while keeping the identity intact. These prompts assume you have attached one or more reference images of your character.
- Same face, new scene:
Using the attached reference images of this person, place them in a sunlit cafe reading a book, keep their face, hair, and skin tone exactly consistent with the references, natural candid pose - Outfit change:
Keep this exact person and face from the references, change their outfit to a tailored navy suit, keep the hairstyle and identity unchanged, studio portrait lighting - Expression change:
Same person as the references, change the expression to a warm genuine laugh, keep the identity, hairstyle, and outfit identical - Multi-angle set:
From the attached reference, generate a clean three-quarter view of the same person with identical features and hair, neutral grey studio background, consistent lighting - Age or era shift:
Keep this person's identity and bone structure from the references, show them in 1970s styling and wardrobe, film-photo look, do not alter the facial identity - Character in action:
Same character from the references, show them mid-stride walking through rain with an umbrella, keep face and hair consistent, cinematic side light
The trick across all of these is to name what must stay fixed — face, hair, skin tone — and only then describe what changes. Identity drift almost always comes from a prompt that forgot to anchor it, so when a likeness slips, the fix is usually to add references and tighten the lock rather than to rewrite the whole scene.
Nano Banana 2 prompts for product shots
Product work is where Nano Banana 2's speed pays off, because you iterate angles and backgrounds quickly before committing. These templates assume a product photo or a reference of the item, and they keep the label and geometry accurate.
- Clean studio:
Place the attached product on a seamless white studio background, soft even lighting, subtle contact shadow under the product, centered, keep the product label and colors exactly as in the reference - Lifestyle scene:
Show the attached product on a marble kitchen counter in morning light, shallow depth of field, a blurred plant in the background, keep the product accurate to the reference - Floating hero:
Render the attached product floating against a soft gradient background, dramatic single key light, crisp reflection beneath, keep the packaging text legible and unchanged - Multi-angle:
From the attached product reference, generate a clean 45-degree three-quarter angle with identical packaging, white background, soft shadow - Color variants:
Keep this product's exact shape and label layout, show it in a matte forest-green colorway, studio lighting, white background - Scale and context:
Place the attached product in a person's hand to show scale, natural skin and daylight, keep the product proportions and label accurate
Because the model can ground details against real references, product geometry and labels hold up better than with most image models — but you should still verify the label text on the final, especially for brand-critical shots that a customer will read closely.
Nano Banana 2 prompts for style transfer
Style transfer applies a look while preserving your content. Nano Banana 2 handles this well when you separate the what from the how. Attach the image you want restyled, and optionally a style reference, then keep the composition locked so only the look changes.
- Reference style match:
Restyle the attached photo to match the color palette and brush texture of the second reference image, keep the original composition and subject placement unchanged - Art movement:
Reinterpret this photo as an art-nouveau illustration, flowing organic lines and muted gold tones, keep the subject recognizable and the composition intact - Film stock:
Give this image the look of expired 35mm film, soft grain, slightly faded colors, gentle halation on highlights, keep all content unchanged - Watercolor:
Convert this landscape into a loose watercolor painting, visible paper texture, soft bleeding edges, preserve the horizon line and main shapes - 3D render look:
Restyle the attached scene as a soft clay 3D render, matte materials, gentle global illumination, keep the layout and proportions the same - Brand palette:
Recolor this image to a brand palette of deep navy, warm cream, and a single coral accent, keep the subject and composition, apply the palette tastefully
The phrase keep the composition is the safety rail here. Without it, style transfer prompts tend to reinvent the framing along with the look, which is rarely what you want when the original layout already worked.
Nano Banana 2 prompts for scene and concept generation
Not every job starts from a photo. When you are generating a scene from scratch, lean into concrete nouns, a clear light direction, and a lens choice, and let the model fill the rest. These prompts need no attachment.
- Establishing scene:
A quiet fishing village at dawn, wooden boats on still water, mist over the hills, soft pink sky, wide 35mm look, gentle reflections, photoreal - Interior:
A cozy reading nook by a rain-streaked window, warm lamp light, a mug on the sill, soft shadows, shallow depth of field, inviting and calm - Concept character:
A weathered desert explorer in practical layered clothing, dust on the fabric, harsh midday sun, full-body, neutral background, grounded and realistic - Food:
A rustic sourdough loaf fresh from the oven on a floured wooden board, steam rising, warm side light, crisp crust detail, appetizing and natural - Abstract background:
A soft gradient background of dusk blues into warm amber, gentle film grain, no subject, suitable as a hero backdrop, 16:9 - Architecture:
A minimalist concrete house at golden hour, large glass facade, long shadows, surrounded by tall grass, architectural photography style
Generation prompts benefit from the model's grounding ability: when you name a real place or object, it can draw on real references to get the details right, which is covered in more depth below.
Nano Banana 2 prompts for text-in-image
Text rendering is a real strength of this model family, and Nano Banana 2 handles legible in-image text far better than older diffusion models. It also supports in-image translation and localization, so you can swap the language of text inside a design without rebuilding it. These prompts cover posters, logos, signage, and labels.
- Poster:
Create a minimalist event poster, large headline reading SUMMER NIGHTS in a bold sans-serif, subheading reading Live Music Fridays, centered layout, deep purple background, keep the spelling exact - Logo concept:
Design a clean wordmark logo for a coffee brand named ROOTED, lowercase rounded sans-serif, a small leaf replacing the dot, monochrome, white background - Signage:
Add a wooden hanging shop sign reading The Corner Bakery in hand-painted serif lettering to the attached storefront photo, match the perspective and lighting, keep spelling exact - Product label:
Generate a skincare bottle label reading CALM, smaller text reading Gentle Daily Serum, clean centered typography, soft sage and white, legible at small size - Localized version:
Take the attached poster and translate the headline text into Spanish, keep the exact layout, fonts, and colors, only change the language of the text - Infographic header:
Create a simple infographic header with the title 2026 Trends in a bold geometric font, three labeled icon placeholders beneath, flat pastel style, accurate spelling
For a text-heavy or brand-final design that must be perfect on the first render, this is the one case where accurate text rendering on the Pro model is worth the slower generation. For drafts and most social graphics, Nano Banana 2 is more than enough.
Negative prompts: telling Nano Banana 2 what to avoid
Nano Banana 2 follows positive instructions well, but a short list of what to avoid sharpens the result, especially for clean commercial work. Fold the exclusions into the same prompt rather than fixing them after the fact.
- Clean product:
Generate the attached product on white, no clutter, no extra props, no text overlay, no reflections on the background - Natural portrait:
Edit this portrait gently, keep natural skin texture, no plastic smoothing, no added makeup, no changes to the eyes - Accurate composite:
Composite the subject onto the new background, match the original lighting, no harsh cutout edges, no mismatched shadows
The pattern is the same each time: state the positive goal, then name the two or three failure modes you have run into before. Because the model renders quickly, you can confirm in one pass whether the exclusions landed and adjust if not. Over a few projects you will collect a short, reusable list of negatives for your own kind of work, and that personal list will do more for your results than any generic one.
Resolution, aspect ratio, and output settings
A prompt is only half the result; the output settings are the other half. Nano Banana 2 can generate at multiple resolutions from a compact 0.5K up to a full 4K, and it supports efficient WebP output. Match the resolution to the job rather than always maxing it out.
- Drafting and iteration: stay at 1K. It renders fastest, which keeps the one-variable-at-a-time loop painless, and it is plenty to judge composition and lighting.
- Social and web: 2K is the sweet spot for crisp on-screen images without wasting render time.
- Print, large displays, or heavy cropping: go to 4K, where the extra detail actually matters and you have room to crop in.
- Aspect ratio: state it in the prompt or the settings — 1:1 for avatars and tiles, 4:5 or 9:16 for vertical social, 16:9 for headers and backdrops. Setting it explicitly avoids awkward recropping later.
A simple workflow: explore at 1K, lock the composition, then re-render the winning prompt at the final resolution. You get the speed while drafting and the detail only where it counts.
Advanced: reference images and web grounding
Two features set Nano Banana 2 apart, and prompts that use them outperform prompts that do not. The first is its support for up to fourteen reference images. The more angles and details you give it of a face, a product, or a style, the more stable the output, so do not stop at one reference when consistency matters.
The second is grounding. The model can look up real-world references to understand what a specific building, landmark, animal species, or object actually looks like before it generates, which means a prompt that names something real tends to come back more accurate than one that leaves it generic. When you want correctness — a recognizable city skyline, a particular breed, a real landmark in the background — name it precisely and let the grounding do the work. When you want pure invention, stay generic and the model will improvise. Knowing which mode you are in is half the skill of prompting this model well.
Troubleshooting: getting the best Nano Banana prompts to work
Even the best Nano Banana prompts sometimes miss on the first try. The fixes are usually small and predictable.
- The model changed too much. Add an explicit lock: keep everything else unchanged, or name the elements that must stay fixed. Edits drift when nothing anchors them.
- The identity shifted. Attach more reference images and state keep the face and hair identical to the references. One reference is rarely enough for a clean likeness.
- The text came out wrong. Put the exact words in capitals or quotes and add keep the spelling exact. If it still misses, move that single design to the Pro model.
- The edit looks pasted-in. Ask the model to match the lighting and color temperature of the original, which is what sells a composite.
- You are iterating slowly. Change one variable at a time — the light, or the background, or the pose — so you can tell which edit did the work. This is also where the speed of Nano Banana 2 helps, since cheap fast renders make one-variable testing painless.
Keep a short note of the phrasings that work for your subjects. A personal prompt library beats a generic one, because your products, faces, and brand have quirks no template can predict.
Run these prompts on PonPon
You can paste every template here into PonPon and switch models without leaving the page. Drop a prompt into the image studio, attach your photo or references, and generate. If a draft is close but needs a brand-final polish, swap from Nano Banana 2 to Nano Banana Pro prompts on the same input — same prompt, higher-fidelity render.
It also helps to have other models a click away for comparison. When a layout depends on flawless typography, running the same brief through OpenAI's GPT Image 2 gives you a second option to choose from.
As you use these, keep refining them into your own set. Save the exact phrasings that work for your faces, products, and brand, note the resolution and reference count that gave the cleanest result, and trim the templates you never reach for. A prompt library is most valuable when it is personal, because the best Nano Banana prompts for your work are the ones you have already proven on your own subjects rather than generic lines from any guide.
The point of a prompt library is not to memorize lines; it is to start from a structure that works, then adapt fast. Copy what fits, change the brackets, and let the model do the rest.