Nano Banana Pro: review, vs version 2 and API access

Google presents Nano Banana Pro as its model for image generation and editing (technically Gemini 3 Pro Image). It outputs 2K and 4K, renders legible text and builds a frame from several references. "Nano Banana" on its own names the whole family, which now has five versions. Below: a version table, how Pro differs from 2, the official addresses and a request through the codexkey API.

What Nano Banana is and which versions exist

Nano Banana is not a product of its own. It is the working name Google uses for the Gemini models that draw and edit images. Searches for "Nano Banana" in any spelling lead to the same family. The versions as of October 29, 2026:

The data comes from the Gemini API image generation docs and the Pro announcement. The name "Nano Banana 2 Pro", which people type into search, does not appear in Google's documentation. Pro and 2 are two different models, not one model with a suffix.

How Nano Banana Pro differs from Nano Banana 2

Pro is built on Gemini 3 Pro. Google describes it as a reasoning engine for professional generation and editing. Version 2 was built for speed. A comparison from Google's data:

The prices in the last row come from Google's price list as of October 29, 2026 (Gemini API pricing page); they are not codexkey's prices. The API has no free tier for these models.

How to choose. For banners, previews and concepts, start with version 2: it is faster and cheaper, and you will need to compare it with Pro on your own prompts anyway. Take Pro when the frame has dense text, an infographic with facts or a scene built from several references. Google's image generation guide recommends Gemini 3 Pro Image specifically for professional assets with text.

Google names the weak spots itself. The Pro page at Google DeepMind says the model can get small details and spelling wrong, infographics and diagrams may contain factual errors, and translating text inside an image comes out with grammar slips. Google asks you to check images together with the text on them. The Nano Banana 2.1 card adds that small text often renders badly and left and right sometimes get mixed up. So if your image needs Cyrillic, test short phrases first, before sending a batch through.

Where to use Nano Banana officially

Nano Banana has no standalone site. A domain with "nanobanana" in it is not necessarily a Google service. The official entry points:

• The Gemini app at gemini.google.com, the Create images tool, with Fast, Thinking and Pro modes.
• Google AI Studio at aistudio.google.com. According to Google's price list, AI Studio is free to use in every supported region.
• The Gemini API: paid, with the key created in AI Studio.

Gemini's image generation page puts it this way: Nano Banana 2 in the app is available at no charge, Nano Banana 2.1 comes with a Google AI plan, and redoing an image with Pro is open to Google AI Pro, Plus and Ultra subscribers. Google does not publish how many images a day you get. Check Settings, then Usage Limits at gemini.google.com.

For a reader in Russia the news here is bad. Russia is not on the Gemini Apps country list, and the list of regions for the Gemini API and AI Studio does not include it either (checked October 29, 2026). We advise against spoofing your country: the account can be flagged for review, and by Google's own documentation, paid plans are open only in some regions. A working route is described below. For free options and limits, read the separate piece: Nano Banana for free.

How to write prompts for Nano Banana Pro

We take three rules from Google's guide. They work better than the templates passed around in chats.

1. Describe the scene in a sentence, not a string of separate words. The more specific the request, the more control you have.
2. For text on the image, give the exact phrase, the font style and the overall design.
3. To edit, upload the image and add what stays untouched. Google's phrase for this is "Keep everything else exactly the same".

Three prompts for a first test:

• Poster: "A poster for a jazz night in a basement, printed on thick paper. A large headline 'Jazz in the Basement' in a sans-serif face at the top, small text 'November 20, 19:00' at the bottom. Warm lamp light, 3:4 format."
• Product: "A photo of an olive oil bottle on a wooden table by a window, soft morning light, shallow depth of field. Only the word 'Olive' is legible on the label."
• Edit: "Replace the background with a bright studio. Leave the bottle, the label and the shadows under it unchanged."

In the results, look at three things: did the phrase come out without extra letters, did the proportions drift, did the label change during the edit. Everything else is secondary.

How to call Nano Banana Pro through the codexkey API

On October 29, 2026 the codexkey catalog has nano-banana, nano-banana-2 and nano-banana-pro. Versions 2.1 and 2 Lite have no IDs of their own. You need a sk_cdx_... key; you can pay by Russian bank card, SBP or crypto, and no Google account is needed. The request goes to POST https://api.codexkey.online/v1/images/generations:

curl https://api.codexkey.online/v1/images/generations \
  -H "Authorization: Bearer sk_cdx_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "nano-banana-pro", "prompt": "Jazz night poster, large headline \"Jazz in the Basement\"", "resolution": "2K"}'

The same in Python, saving the file:

from pathlib import Path
from urllib.parse import urlparse

import requests

API_KEY = "sk_cdx_YOUR_KEY"

response = requests.post(
    "https://api.codexkey.online/v1/images/generations",
    headers={"Authorization": f"Bearer {API_KEY}"},
    json={
        "model": "nano-banana-pro",
        "prompt": "Jazz night poster, large headline \"Jazz in the Basement\"",
        "resolution": "2K",
    },
    timeout=300,
)
response.raise_for_status()

image_url = response.json()["data"][0]["url"]
suffix = Path(urlparse(image_url).path).suffix or ".png"
Path(f"poster{suffix}").write_bytes(requests.get(image_url, timeout=60).content)

The link in data[].url is temporary, so save the file right away. The resolution parameter takes 1K, 2K or 4K, and size sets a size such as 1024x1024. The allowed values for each model are listed on its page in the catalog. Editing from a reference goes to POST /v1/images/edits; details are in the images and video reference. You are charged only for a successful result.

nano-banana-2 has a lightweight mode: add "tier": "lite" and drop resolution. Whether it matches Google's Nano Banana 2 Lite, the codexkey documentation does not say, so compare the output on your own tasks.

How not to overpay: Cheap and lite

The public codexkey price list has no image prices; they are on each model's page. The channel setup is documented, though, and it affects your choice.

• nano-banana-pro is sold in the Cheap channel too: noticeably cheaper, but generation fails more often. nano-banana-2 is not in Cheap and always goes through Premium.
• If the key is in cheap mode and generation fails, you get a 503 model_unavailable error and nothing is charged. In auto_cheap mode the request goes to Premium and is billed at the Premium price.
• The channel that made the image comes back in the channel field of the response. How channels work is covered in the Premium and Cheap guide.

The setup we would use for volume: drafts on nano-banana-2 or lite, final frames on nano-banana-pro at 2K. For bulk generation with a retry on error, Cheap makes sense. For one urgent poster we would not risk it: a rerun will blow the deadline harder than the price difference saves.

If you are choosing between Google and OpenAI, there is a scenario-by-scenario comparison in GPT Image vs Nano Banana: how to choose an image API, and a separate review of the competitor here: GPT Image 2 and 2.5. What the free access gives and where it ends is covered in Nano Banana for free. For an open model you can run yourself, there is the Qwen Image 2.1 review.

FAQ

What is Nano Banana Pro in plain terms?

It is Google's model for generating and editing images, built on Gemini 3 Pro. It came out on November 20, 2025, outputs 2K and 4K, renders legible text and builds a frame from several references. A search for "nano banana" usually means any model in this family.

Which should I pick: Nano Banana Pro or Nano Banana 2?

Version 2 is faster and, on the Gemini API price list, half the price at 1K, while Pro suits dense text, infographics and complex references better. Start with version 2 and move to Pro if the result misses your criteria.

Does Nano Banana have an official website?

No standalone site. The models are available in the Gemini app, Google AI Studio and through the Gemini API, and the descriptions live on Google and Google DeepMind pages. Domains with nanobanana in the name are not necessarily tied to Google.

Does Nano Banana work in Russia?

Not directly: Russia is not on the Gemini Apps country list or on the region list for the Gemini API and AI Studio as of October 29, 2026. Through the codexkey API, the nano-banana, nano-banana-2 and nano-banana-pro models can be called without a Google account.

How much does Nano Banana Pro cost?

In the Gemini API Google charges $0.134 per image at 1K and 2K and $0.24 at 4K (price list of October 29, 2026). For codexkey, see the model's page in the catalog: the public price list has no image prices, and the ruble rate floats.

How do I call Nano Banana from Python through the API?

Send a POST to https://api.codexkey.online/v1/images/generations with an Authorization header and the fields model, prompt and resolution, then download the file from the link in data[].url. A full example is given earlier in the article.