Nano Banana Pro API

An image costs $0.027, and resolution does not change that number. 1K and 4K are billed the same. Everywhere else we checked prices in August, 4K carries a double rate.

What the model is for

Nano Banana Pro is Google's image model. We carry it as one SKU with two things to set, the aspect ratio of the frame and the resolution, anywhere from 1K to 4K. It also accepts reference images, so a request along the lines of "same person, now in a forest" is squarely its job.

That shapes what it gets used for. Aspect ratio is explicit, so the image arrives already fitting the layout and nobody has to crop it afterwards. References go in when you need a series holding one character or one style. And 4K is for work headed to print or a large screen.

Price

Every image that comes back costs $0.027. Ask for 4K instead of 1K, a wide frame instead of a square, attach three references, and the number stays where it was. A provider-side failure returns a zero in cost_usd and leaves the balance alone.

What the same model costs at other aggregators

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

The cheapest of them charges more than three times our rate, the rest five and up. At 4K the gap widens further, because the double tier that kicks in for them does not exist here.

So that this does not read one-sided. fal.ai and replicate carry a thousand-plus models against our fifty, and fal.ai holds SOC 2 on top of that. If what you need is a hundred different generators under one bill, a price gap on a single model will not outweigh it. If what you need is this model, per-second video next to it, and no subscription, the arithmetic above is yours to run.

How to call it

Same endpoint as every other image model, POST /v1/images/generations.

curl https://api.teamtoken.store/v1/images/generations \
  -H "Authorization: Bearer $TEAMTOKEN_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nano-banana-pro",
    "prompt": "bakery window early morning, light from the street, film grain",
    "aspect_ratio": "16:9",
    "resolution": "4K"
  }'

A reference goes in the image field, several go in an images array. URLs, data-URLs and bare base64 all work.

curl https://api.teamtoken.store/v1/images/generations \
  -H "Authorization: Bearer $TEAMTOKEN_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nano-banana-pro",
    "prompt": "same person, in a forest, golden hour",
    "images": ["https://example.com/ref-1.jpg", "https://example.com/ref-2.jpg"],
    "aspect_ratio": "4:3"
  }'

The response carries b64_json plus a cost_usd field with what was actually charged.

What you can tune

References come with one rule that trips people up. Within a single request use either base64 or links, not a mix of both. JPG and PNG only.

Unknown fields do not break the request, they get dropped before it reaches the provider. Convenient, right up until you send size instead of resolution and start wondering why the image keeps coming back smaller than you asked.

The response is synchronous and carries base64, one image per call: n is validated but never forwarded to the provider. If generation misses the waiting window, you get an HTTP 202 with a job id instead, and the result is collected later from GET /v1/images/jobs/{id}. Your money is on hold at that point and not yet charged.

Nano Banana Pro or Nano Banana 2

Both cost $0.027, take the same fields and hit the same endpoint. Price and interface cannot tell them apart, which leaves the images themselves as the deciding vote.

What is known about their origins. Aggregators that disclose the underlying model id put Pro on the pro branch of Google's image family and the other one on the flash branch. Flash branches tend to be faster and simpler, but we have no benchmark of our own, and somebody else's naming is not a measurement.

The practical answer is dull and it works: run one prompt through both, that is $0.054 spent, and keep whichever landed closer.

One key for images, video and text

The key does not split by model type and the balances do not drift apart. The same token calls video on /v1/videos, billed per second of finished clip and starting at $0.0067 with grok-video-480p, and text models on /v1/chat/completions. One bill for all of it, with the ceiling set per key.

No gallery here, but there is something to test

The absence of a gallery is deliberate. This page's main claim is about the invoice rather than the picture, and two calls settle it.

The whole set costs $0.05.

Questions we get