Nano Banana Logo Design: Prompts, Models and Cost (2026)

A Nano Banana logo is a logo concept generated with Google’s Nano Banana image models, the Gemini-based family that produces clean, natural-looking images and handles text far better than older image generators. This guide shows which model to use at each stage, how to write prompts that give you a usable mark instead of a pretty poster, what it costs per image, and how to turn the output into the PNG and SVG files a real brand needs.

What people mean by “Nano Banana logo”

The search covers two different things, so it helps to separate them.

Most people want to design a logo with Nano Banana. Nano Banana is the nickname for Google’s Gemini image model (officially Gemini 2.5 Flash Image). It was followed by Nano Banana Pro (officially Gemini 3 Pro Image) and a second generation, Nano Banana 2. A text-to-image model takes a plain-language description, called a prompt, and returns an image that matches it. Because these models render letters accurately and follow instructions closely, they became a popular way to sketch logo ideas in minutes.

A smaller group wants the official Nano Banana logo itself, the banana branding Google uses for the model, usually as a PNG. That artwork belongs to Google. If you need it for an article or presentation, get it from Google’s own brand resources and follow their guidelines. Do not put it into your own product’s logo or app icon.

The rest of this article covers the first case: making your own Nano Banana logo that you can actually ship.

Which Nano Banana model is best for logos?

There are three Nano Banana models worth knowing for logo work, plus two non-Google models that fill gaps the family leaves. Prices below are what each image costs on Unicorn API, where one credit costs $0.01 on a monthly plan.

Model	Best logo use	Reference images	Price per image
Nano Banana	Fast concept exploration and quick edits	Up to 3	4 credits
Nano Banana 2 Lite	Cheap large batches at 1K, mood-board matching	Up to 10	4 credits
Nano Banana Pro	Final wordmarks with exact spelling, 2K or 4K output	Up to 8	18 credits (1K, 2K), 24 credits (4K)
GPT Image 1.5	Transparent-background versions, legible text	Up to 4	3 (low), 4 (medium), 22 (high) credits
Flux 3 Klein Edit Image	Very cheap iterative edits of a chosen mark	Via image URLs	1.6 credits at 1K, rounded up per request

When to use each

• Nano Banana is the sensible default for the first round. It is quick, it is cheap, and it produces flat, natural shapes without much fuss.
• Nano Banana 2 Lite costs the same per image but accepts up to 10 reference images. If you have a folder of brands you like, packaging photos or a hand sketch, feed several in and ask for a mark in that spirit. It outputs 1K images, which is plenty for judging a concept.
• Nano Banana Pro is the one to use when the name has to be spelled correctly, every time, and when you want a high-resolution master file. It costs about 4.5x more than the base model, so save it for finalists.
• GPT Image 1.5 is not a Nano Banana model, but it supports a transparent background, which none of the Nano Banana models list as an option. That makes it useful for the final export.

How to write a Nano Banana logo prompt

Image models are trained mostly on photos and illustrations, so a vague prompt pulls them toward rich, detailed scenes. A logo needs the opposite: few shapes, few colours, a lot of empty space. A good Nano Banana logo prompt states those constraints explicitly.

The seven-part prompt framework

1. Brand name in quotation marks. Quotes tell the model these exact characters must appear. Example: "Fernhill".
2. Logo type. Say whether you want a wordmark (the name styled as type), a symbol (an icon on its own) or a combination mark (symbol plus name).
3. One symbol idea. Describe a single concept, such as “a fern leaf forming the shape of a coffee bean”. Two ideas usually produce a cluttered mark.
4. Style. “Flat vector logo, geometric, consistent line weight” beats “beautiful modern logo”.
5. Colours. Name two or three: “deep forest green and cream”. Include hex codes if you have brand colours.
6. Background and framing. “Centred on a plain white background, generous padding.” This makes cleanup and vectorising far easier.
7. What to avoid. “No gradients, no shadows, no 3D, no photographic texture, no extra text.”

A complete example prompt

Combination mark logo for a specialty coffee roaster named "Fernhill".
Symbol: a single fern leaf whose outline also reads as a coffee bean.
Wordmark: "Fernhill" in a clean geometric sans-serif, lowercase, to the right of the symbol.
Style: flat vector logo, two colours only, deep forest green (#1F4D3A) and cream (#F3EBDD).
Centred on a plain white background with generous padding.
No gradients, no shadows, no 3D effects, no mockup, no extra words.

Prompt variations worth trying

• Monogram: “A monogram logo using the letters ‘FH’, interlocking, single colour.”
• Emblem: “A circular badge logo with ‘Fernhill Coffee’ around the edge and a fern leaf in the centre.”
• Negative space: “A coffee cup silhouette where the steam forms a hidden leaf in the negative space.”
• App icon: “The Fernhill fern-bean symbol only, no text, inside a rounded square, readable at small sizes.”

Short prompts work well. If an output is close but wrong, change one instruction at a time so you know what made the difference.

Generate a Nano Banana logo with the API

You can try every model by hand in the browser playground, which is the fastest way to tune a prompt. Once you want batches, consistent settings or an automated pipeline, call the API. Image generation on Unicorn API runs as a job: you submit a request with POST /v1/jobs, poll GET /v1/jobs/{id} until the status is "succeeded", then read the image URL from outputs[0].url.

First create an API key and export it as UNICORN_API_KEY.

curl: one concept with Nano Banana

curl https://api.unicornapi.net/v1/jobs \
  -H "Authorization: Bearer $UNICORN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nano-banana",
    "input": {
      "prompt": "Flat vector combination mark logo for a coffee roaster named \"Fernhill\": a fern leaf shaped like a coffee bean, forest green and cream, plain white background, no gradients, no shadows",
      "aspect_ratio": "1:1",
      "n": 4
    }
  }'

Setting n to 4 returns four variations in one job. The price is per output, so this costs 4 x 4 = 16 credits.

Python: explore, then render the finalist with Nano Banana Pro

import os
import time
import requests

BASE = "https://api.unicornapi.net"
HEADERS = {
    "Authorization": f"Bearer {os.environ['UNICORN_API_KEY']}",
    "Content-Type": "application/json",
}

PROMPT = (
    'Combination mark logo for a specialty coffee roaster named "Fernhill". '
    "Symbol: a single fern leaf whose outline also reads as a coffee bean. "
    'Wordmark: "Fernhill" in a clean geometric sans-serif, lowercase. '
    "Flat vector style, two colours only: deep forest green and cream. "
    "Centred on a plain white background with generous padding. "
    "No gradients, no shadows, no 3D, no extra words."
)


def run_job(model, input_fields, timeout=300):
    """Submit a job and poll until it succeeds. Returns the job JSON."""
    resp = requests.post(
        f"{BASE}/v1/jobs",
        headers=HEADERS,
        json={"model": model, "input": input_fields},
    )
    resp.raise_for_status()
    job = resp.json()

    deadline = time.time() + timeout
    while job.get("status") != "succeeded":
        if job.get("status") in ("failed", "error", "canceled"):
            raise RuntimeError(f"Job {job.get('id')} ended with status {job.get('status')}")
        if time.time() > deadline:
            raise TimeoutError(f"Job {job.get('id')} did not finish in time")
        time.sleep(3)
        poll = requests.get(f"{BASE}/v1/jobs/{job['id']}", headers=HEADERS)
        poll.raise_for_status()
        job = poll.json()
    return job


def download(url, path):
    img = requests.get(url)
    img.raise_for_status()
    with open(path, "wb") as f:
        f.write(img.content)


# Round 1: four cheap concepts with Nano Banana (4 x 4 = 16 credits)
concepts = run_job("nano-banana", {"prompt": PROMPT, "aspect_ratio": "1:1", "n": 4})
for i, out in enumerate(concepts["outputs"]):
    download(out["url"], f"concept_{i + 1}.png")
    print("concept", i + 1, out["url"])

# Round 2: re-render the favourite at 2K with Nano Banana Pro (18 credits),
# passing the chosen concept as a reference image.
favourite_url = concepts["outputs"][0]["url"]
final = run_job(
    "nano-banana-pro",
    {
        "prompt": PROMPT + " Keep the composition of the reference image; make the lettering crisp and exact.",
        "ref_images": [favourite_url],
        "resolution": "2K",
        "aspect_ratio": "1:1",
    },
)
download(final["outputs"][0]["url"], "fernhill_final_2k.png")
print("final", final["outputs"][0]["url"])

Two practical notes. First, the exact price of any request is available before you run it through POST /v1/models/price, which is useful when a script builds requests dynamically; the request format is in the API documentation. Second, if you work inside Cursor or Claude Code, you can call the same models from your editor through the MCP integration and iterate on prompts without leaving your project.

Worked example: from concepts to a finished mark

Here is a realistic run for the Fernhill brand, with the credit cost of each step.

1. Explore broadly. Run the base prompt plus the monogram, emblem and negative-space variations through Nano Banana, four images each. That is 16 images at 4 credits, 64 credits. You will discard most of them, and that is the point: concepts are cheap.
2. Match a mood board. Take the three directions you liked and run them through Nano Banana 2 Lite with up to 10 reference images of packaging and type you admire. Eight images at 4 credits is 32 credits.
3. Render finalists. Send the best three concepts to Nano Banana Pro at 2K as reference images, with the lettering instruction added. Three images at 18 credits is 54 credits.
4. Make small edits. Use Flux 3 Klein Edit Image for tweaks such as “make the leaf stem thinner” or “swap green for #2E6B4F”. At 1K each edit is 1.6 credits, rounded up to 2 for a single image, so ten edits is about 20 credits.
5. Export a transparent version. Render the chosen design once with GPT Image 1.5 at high quality with a transparent background, 22 credits.

Total: 64 + 32 + 54 + 20 + 22 = 192 credits. On a monthly plan that is $1.92; with top-up credits at $0.015 each it is $2.88. Either way, credits never block you mid-project: plans start with Starter at $25 per month, and top-ups never expire. See pricing for the plan details.

The money is rarely the constraint. The real cost is your time judging outputs, which is why the workflow front-loads cheap, fast models and only spends Pro credits on designs you would actually ship.

Transparent PNG, SVG and other file formats

A finished brand needs more than one image. It typically needs a full-colour version, a single-colour version, a version for dark backgrounds, an icon on its own, and a vector master that scales to a billboard. Every Nano Banana logo starts as a raster image, meaning a fixed grid of pixels, so plan the conversion.

Getting a transparent PNG

• Generate on plain white, then remove the background in any image editor. Flat designs on white make this a one-click job.
• Use GPT Image 1.5 with a transparent background for the final export. Set the background input field; check the docs for accepted values.

curl https://api.unicornapi.net/v1/jobs \
  -H "Authorization: Bearer $UNICORN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-1.5",
    "input": {
      "prompt": "Flat vector logo: a fern leaf shaped like a coffee bean beside the lowercase word \"fernhill\", forest green, no background",
      "quality": "high",
      "background": "transparent",
      "aspect_ratio": "1:1"
    }
  }'

Getting an SVG

No image model in this list outputs vectors. To get an SVG, trace the cleanest high-resolution PNG with a vector tracing tool, then open the result in a vector editor and fix it by hand: merge stray nodes, straighten lines that should be straight, and replace AI-drawn lettering with a real font if the type looks slightly off. Rendering the source at 2K or 4K with Nano Banana Pro gives the tracer more detail to work with, and 4K costs 24 credits instead of 18.

Testing at real sizes

Shrink the mark to 32 pixels wide, the size of a browser tab icon. If the symbol turns into a smudge, simplify it. Also check it in pure black on white. A logo that only works in colour will cause problems on receipts, embroidery and fax-era paperwork that still exists.

Common Nano Banana logo mistakes

• Asking for too much. “A coffee bean, a leaf, a mountain and a sunrise” gives you an illustration, not a logo. Pick one idea.
• Accepting gradients and shadows. They look rich on screen and break in print, embroidery and single-colour use. Ban them in the prompt.
• Not checking the spelling. Nano Banana Pro is strong at text, but read every letter anyway, especially on long or invented names. A missing letter in a final file is an expensive surprise.
• Requesting a grid of logos in one image. “Show me nine logo options” produces tiny, low-detail marks. Use n to get separate full-size outputs instead.
• Skipping the vector step. A 1K PNG is fine for a website header and useless for signage. Trace and clean it.
• Copying an existing brand. Reference images guide style; they should not be used to reproduce another company’s mark. Search trademark databases for your name and symbol before launch.
• Ignoring ownership rules. In the US, copyright generally requires human authorship, so a purely generated mark may have limited protection. Having a designer redraw and refine the chosen concept strengthens your position, and trademark registration protects the mark in commerce regardless of how it was made.
• Using Pro for everything. At 18 credits per image versus 4, running every exploratory prompt through Nano Banana Pro multiplies your bill for no gain. Explore cheap, finish premium.

If you want to compare how other models handle the same brief, the full model catalog lists every image model available on one key, including Flux 3 Klein Text to Image at about 2 credits per 1K image for very wide, cheap exploration. Run your Fernhill-style prompt through two or three of them, put the outputs side by side, and keep whichever direction holds up at 32 pixels in black and white. That test, more than the model name, decides whether your Nano Banana logo becomes a real brand.