Nano Banana Examples: 10 Prompts With API Code (2026)

The most useful nano banana examples are not the viral ones; they are the repeatable prompt patterns that turn a plain request into a reliable result, whether you are generating a product shot from scratch or editing a photo you already have. This post collects ten of them, explains which version of Google’s model each one needs, shows what each costs, and gives you working code to run them in bulk.

Every prompt below is written so you can paste it, swap the subject, and get a usable image. Where a prompt depends on reference images (photos you upload so the model can copy a face, product or room), that is called out.

What is Nano Banana, and which version are these examples for?

“Nano Banana” is the nickname for Google’s Gemini image models. It started as a codename and stuck, so today it covers a small family rather than one model. Searches for google nano banana, gemini nano banana and nano banana ai all point to the same set of models:

• Nano Banana (officially Gemini 2.5 Flash Image), released August 2025. Quick, natural-looking images and edits with up to 3 reference images.
• Nano Banana Pro (officially Gemini 3 Pro Image), released November 2025. High-fidelity images with accurate text rendering at 1K, 2K or 4K, with up to 8 reference images.
• Nano Banana 2 Lite, released July 2026. The cheapest Gemini image model: fast 1K images and edits with up to 10 reference images. If you searched “nano banana 2.0”, this is the second-generation family it belongs to.

All three do both text-to-image (a text-to-image model takes a written prompt and produces a matching picture) and image editing, where you supply one or more photos and describe the change. The differences that matter in practice are text quality, resolution, how many references you can pass, and price.

Model	Best at	Max reference images	Resolution	Credits per image
Nano Banana	Natural photos, quick edits	3	Set via image_size	4
Nano Banana Pro	Text, layouts, 4K finals	8	1K, 2K, 4K	18 (1K/2K), 24 (4K)
Nano Banana 2 Lite	Cheap volume, many references	10	1K	4

On Unicorn API, monthly plans give credits at $0.01 each, so 4 credits is $0.04 and 18 credits is $0.18. Top-up credits cost $0.015 each and never expire. Full details are on the pricing page.

Nano banana examples for text-to-image

These first four nano banana examples start from nothing but a prompt. The pattern that works across all of them is: subject, setting, light, camera or style, then hard constraints. Constraints go last and are stated plainly (“no text other than...”, “plain white background”).

1. Product shot on a seamless backdrop

A matte black stainless steel water bottle standing upright on a seamless
warm grey backdrop, soft studio lighting from the upper left, subtle shadow
on the floor, 85mm product photography, sharp focus on the logo area,
no props, no text.

Model: Nano Banana at 1:1, 4 credits. The base model’s strength is natural-looking light, which is exactly what catalog images need. “No props, no text” stops the model from decorating the scene.

2. Poster with exact headline text

A vintage jazz club poster, deep navy background with a gold saxophone
illustration. Headline at the top in bold art-deco lettering:
"BLUE NOTE FRIDAY". Below it in smaller text: "Doors 8pm, Live Quartet".
Print texture, slight paper grain. No other text anywhere.

Model: Nano Banana Pro at 2K, 9:16, 18 credits. Put every word you want rendered inside quotation marks and say where it goes. Accurate text rendering is the main reason to pay for Pro, and posters, book covers, packaging concepts and landing page hero images all fall into this category.

3. Labelled infographic or diagram

A clean flat-design infographic explaining how coffee is brewed with a
pour-over. Four numbered panels left to right: "1 Grind", "2 Bloom",
"3 Pour", "4 Serve". Each panel has a simple icon. White background,
two accent colours: teal and orange. Large readable sans-serif labels.

Model: Nano Banana Pro at 4K, 16:9, 24 credits. Keep labels short (one or two words each); the more text you ask for, the more chances for a slip. Use 4K only if the image will be printed or shown large.

4. One concept, several aspect ratios

A cosy reading nook by a rain-streaked window, a mug of tea on the sill,
warm lamp light, autumn evening, cinematic photograph, shallow depth of field.

Model: Nano Banana 2 Lite, 4 credits per image. Run the same prompt at 9:16 for stories, 3:2 for a blog header and 1:1 for a feed post. 2 Lite supports 3:2 and 2:3, which the other two Nano Banana versions do not list, and it is cheap enough to generate several variations per format.

Nano banana examples for photo editing

Editing is where Nano Banana built its reputation. You pass reference images in the ref_images field and describe the change in plain language. The key habit: say what should change and, just as clearly, what must stay the same.

5. Fix lighting on an existing photo

Relight this portrait as if it were taken during golden hour: warm low
sunlight from the right, soft shadows on the left side of the face.
Keep the person's face, expression, hair and clothing exactly the same.
Do not change the background composition.

Model: Nano Banana with one reference image, 4 credits. Relighting is one of the most practical edits for real-estate, food and portrait photos shot in bad conditions.

6. Recolor a product without touching anything else

Change the sneaker's upper from white to forest green. Keep the sole,
laces, logo, stitching, shadows and background identical.

Model: Nano Banana, 4 credits. This one pattern can replace a whole colourway photo shoot. Run it once per colour and compare the results side by side for drift in the logo or stitching.

7. Add objects to a room (interior redesign)

Redesign this living room in a Japandi style: add a low oak coffee table,
a beige linen sofa against the back wall, and a large paper floor lamp in
the corner. Keep the walls, windows, flooring and camera angle unchanged.

Model: Nano Banana Pro with the room photo plus up to 7 product photos as extra references, 18 credits at 2K. Passing real furniture photos as references is how interior and e-commerce teams get the actual product into the scene rather than a lookalike.

8. Photo to collectible figurine with packaging

Turn the person in this photo into a 1/7 scale collectible figure standing
on a round clear acrylic base on a desk. Behind it, a retail toy box with
the same character printed on the front and the name "MAYA" on the box.
Realistic studio lighting, product photography style.

Model: Nano Banana Pro, 18 credits. This was one of the most-shared trends after launch. Pro is the better choice here because the box text needs to be legible.

9. Same character across several scenes

Using the character in the reference images, show her reading a map at a
busy train station. Same face, hairstyle, red raincoat and backpack as the
references. Documentary photo style, overcast daylight.

Model: Nano Banana 2 Lite with up to 10 references, 4 credits per image. More references from different angles give the model a firmer idea of the character, which is what storyboards, comics and ad campaigns need. If the finals must be 4K, switch to Pro with up to 8 references.

10. Sketch or map to photograph

Turn this hand-drawn floor plan sketch into a photorealistic top-down
render of the apartment. Wooden floors, white walls, furniture placed
where the sketch shows it. Keep the room proportions from the sketch.

Model: Nano Banana Pro, 18 credits. The same idea works for turning a map location into a street-level scene, or a napkin wireframe into a website mockup. The phrase “keep the proportions from the sketch” does most of the work.

How to run these nano banana examples through the API

The free Gemini app is fine for trying a prompt once. When you want to run nano banana examples across 200 products, or put them behind a button in your own app, you need an API. On Unicorn API every image model uses the same pattern: create a job with POST /v1/jobs, poll GET /v1/jobs/{id} until status is "succeeded", then download job.outputs[0].url.

First, create an API key and export it as UNICORN_API_KEY. The quickest test is a single curl call:

curl -X POST https://api.unicornapi.net/v1/jobs \
  -H "Authorization: Bearer $UNICORN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"nano-banana","input":{"prompt":"A matte black water bottle on a seamless warm grey backdrop, soft studio light, no text","aspect_ratio":"1:1"}}'

That returns a job object with an id. Here is a complete Python script that runs a text-to-image prompt with Pro and an edit with the base model, polling each to completion:

import os
import time
import requests

API = "https://api.unicornapi.net"
HEADERS = {"Authorization": f"Bearer {os.environ['UNICORN_API_KEY']}"}


def run_job(model, input_fields, poll_seconds=3, timeout=300):
    resp = requests.post(
        f"{API}/v1/jobs",
        headers=HEADERS,
        json={"model": model, "input": input_fields},
    )
    resp.raise_for_status()
    job = resp.json()

    deadline = time.time() + timeout
    while job["status"] != "succeeded":
        if job["status"] in ("failed", "canceled"):
            raise RuntimeError(f"Job {job['id']} ended with status {job['status']}")
        if time.time() > deadline:
            raise TimeoutError(f"Job {job['id']} still {job['status']}")
        time.sleep(poll_seconds)
        resp = requests.get(f"{API}/v1/jobs/{job['id']}", headers=HEADERS)
        resp.raise_for_status()
        job = resp.json()

    return job["outputs"][0]["url"]


# Example 2: poster with exact text, Nano Banana Pro at 2K
poster_url = run_job("nano-banana-pro", {
    "prompt": (
        'A vintage jazz club poster, deep navy background with a gold saxophone '
        'illustration. Headline at the top in bold art-deco lettering: '
        '"BLUE NOTE FRIDAY". Below it in smaller text: "Doors 8pm, Live Quartet". '
        'No other text anywhere.'
    ),
    "resolution": "2K",
    "aspect_ratio": "9:16",
})
print("Poster:", poster_url)

# Example 6: recolor a product photo, Nano Banana with one reference
recolor_url = run_job("nano-banana", {
    "prompt": (
        "Change the sneaker's upper from white to forest green. Keep the sole, "
        "laces, logo, stitching, shadows and background identical."
    ),
    "ref_images": ["https://example.com/white-sneaker.jpg"],
    "aspect_ratio": "1:1",
})
print("Recolor:", recolor_url)

A few practical notes:

• Reference images go in ref_images. The API documentation lists the accepted formats; publicly reachable image URLs are the simplest option.
• Set n to get several variations from one request. Price is per output, so n: 4 on the base model is 16 credits.
• To switch models, change one string. The same script runs "nano-banana-2-lite", or a non-Google model such as GPT Image 1.5, without new keys or a new bill.
• Before a large batch, call POST /v1/models/price to get the exact cost of a request.

If you would rather test prompts by hand first, the image playground runs the same models in the browser. If you work inside Cursor or Claude Code, the agent integration lets your coding assistant call these models directly.

Which Nano Banana version should you use, and what does it cost?

The honest answer for most work: draft cheap, finish on Pro only when you need to. Here is how the examples above map to models, with costs on a monthly plan at $0.01 per credit.

Task	Recommended model	Credits	Cost on a plan
Product shots, relighting, recoloring	Nano Banana	4	$0.04
Multi-format social sets, character consistency	Nano Banana 2 Lite	4	$0.04
Posters, packaging, anything with text	Nano Banana Pro 1K/2K	18	$0.18
Print-size infographics and finals	Nano Banana Pro 4K	24	$0.24

A worked example: a 40-product colour launch

Say you sell 10 sneakers, each in 4 new colours, and need one image per combination.

1. Recolor all 40 with Nano Banana: 40 × 4 = 160 credits.
2. About a quarter will drift (wrong stitching, shifted logo). Rerun 10 with a tighter prompt: 10 × 4 = 40 credits.
3. Produce 4 hero images with text overlays on Nano Banana Pro at 4K: 4 × 24 = 96 credits.

Total: 296 credits, which is $2.96 on any monthly plan, or $4.44 if you paid with top-up credits at $0.015 each. The Starter plan at $25 a month covers this many times over.

When to look beyond Nano Banana

Nano Banana is not always the cheapest or best fit. For very high volumes of simple images, Flux 3 Klein Text to Image costs 1.6 credits at 1K, and Flux 3 Klein Edit Image handles edits at the same price. GPT Image 1.5 offers an optional transparent background, which saves a cut-out step for logos and stickers, at 3 credits on low quality or 4 on medium. Because all of these sit behind the same API key, you can run the same prompt on two models and keep whichever wins. The full model catalog lists everything available.

Common mistakes with Nano Banana prompts

Most failed nano banana examples you see online fail for the same handful of reasons.

• Not saying what should stay the same. “Make the sofa green” can change the walls too. Always add a “keep X, Y and Z unchanged” clause to edits.
• Asking the base model for lots of text. Short labels sometimes work on Nano Banana, but paragraphs and exact brand names belong on Pro.
• Using 4K for drafts. Pro costs the same at 1K and 2K, and 4K adds a third on top. Iterate at 2K and render 4K only for the final.
• Low-quality references. Blurry, dark or heavily cropped reference photos produce blurry, dark results. Send the best source you have.
• Too few angles for consistency. One reference photo of a character is a guess; four to six from different angles is a likeness. Nano Banana 2 Lite accepting 10 is useful here.
• Stacking contradictory styles. “Photorealistic watercolor anime render” gives the model nothing to commit to. Pick one style per prompt.
• Running one huge batch without a test. Run five images, fix the prompt, then run the batch. Checking the price endpoint first avoids surprises.

One more subtle one: copying viral prompts word for word. They were tuned for a specific photo. Copy the structure, not the wording.

Where to find more nano banana examples

Once you have the patterns above, you will want more ideas. Searches like “awesome nano banana” and “nano banana github” lead to community repositories that pair hundreds of prompts with their outputs, and prompt galleries let you copy prompts with one click. They are useful for spotting categories you had not considered: museum-style labels, ID-badge mockups, recipe cards, before-and-after renovations, historical photo restoration.

Read them critically. Galleries show winners, not the five attempts before them, and many images were made with older model versions. The quickest way to judge a prompt is to run it yourself on two versions and compare: in the playground for a handful, or with the script above for a batch. Treat every gallery entry as one of many nano banana examples to adapt, not a guarantee.

If you are building a product around these, a few more related guides may help: our roundup of the best AI image upscalers covers taking 1K Nano Banana 2 Lite drafts up to print size, and Seedream 5 Lite is another image model worth comparing for product work. To animate a finished still, see Runway image to video.

The short version: start with the ten nano banana examples here, run them on the cheapest model that handles the job, and move to Nano Banana Pro only when text, 4K or many references demand it. One key, one bill, and you can switch models by changing a single string.