Nano Banana 2 API

An image costs $0.027 at any resolution, 4K included. Nano Banana Pro costs exactly the same, which is the first thing worth knowing if you are picking between the two.

Nano Banana 2 or Pro, when the price is identical

Most people land here with the choice half made already. Price will not finish it for you: the two models cost the same, take the same fields and go to the same endpoint. No request parameter hints at which one suits your work better.

Here is what is known from the outside. Aggregators that publish the underlying model id put Pro on the pro branch of Google's image family and this one on the flash branch. Flash branches have a reputation for being quicker and less fussy about fine detail. We have run no benchmark of our own, and repeating somebody else's naming as though it were a test result is not something we will do.

So the advice is prosaic. Take a prompt you actually need, send it through both, spend $0.054 and a minute of waiting. The answer you get is about your work rather than about a benchmark somebody else ran. Then keep the one that landed closer and stop thinking about it.

Price

$0.027 for every image that comes back. Resolution, aspect ratio and attached references leave the number alone, because the price attaches to the call. When the provider fails, the response carries a zero cost_usd and the balance stays put.

The same model priced elsewhere

Prices read on 22.08.2026 from each service's own public pages.

We come in under the cheapest of them, though the margin here is modest: one and a half times against replicate, four times against the top of the range. On Pro the spread is far wider, so if you are sizing us up against the market, read both pages rather than this one alone.

How to call it

POST /v1/images/generations, using the same key your text models use.

curl https://api.teamtoken.store/v1/images/generations \
  -H "Authorization: Bearer $TEAMTOKEN_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nano-banana-2",
    "prompt": "ceramic mug on a windowsill, morning light, soft shadows",
    "aspect_ratio": "4:3",
    "resolution": "2K"
  }'

Reference images ride in the image field or an images array. URLs, data-URLs and bare base64 all work, up to four of them, JPG and PNG. Inside a single request, do not mix base64 with links.

Parameters

About n, plainly, so you do not lose an evening to it. We range-check the field but never forward it, and a call returns one image. Ten images means ten calls, ideally in parallel. Billing counts what actually came back.

Unknown fields raise no error, they are quietly dropped. Send size instead of resolution and the image arrives at 1K without a word of warning.

Waiting runs to tens of seconds and the connection stays open throughout. If the provider misses the 150-second mark, an HTTP 202 with a job id arrives instead, and the image is collected afterwards from GET /v1/images/jobs/{id}.

Images, video and text on one key

There is no split between a media account and a text account. The same key calls video on /v1/videos, billed per second of finished clip, and text models on /v1/chat/completions. The balance is shared, while the spend ceiling is set on the individual key rather than on the whole account.

No gallery here, but there is something to test

This whole page comes down to the choice being settled by the image rather than the price list. So the test is an image, on your prompt instead of ours.

The whole set costs $0.05.

Questions we get