Nano Banana Pro API: Slow, $0.075, and Worth It Anyway
Someone types "fast and affordable Nano Banana Pro API" into a search box and we have to argue with the first word immediately. Nano Banana Pro is not fast. A request lands in roughly 30 seconds through our pipeline, where the flash tiers finish in two or three.
That gap is not queueing and it is not us being slow. It is the model working, and it is the entire reason the tier exists.
We sell Google's gemini-3-pro-image at $0.075 per request as of our 26 August 2026 catalog check. Notice the model ID, because a lot of coverage calls this thing "Gemini 3.0 Pro" and no such model exists. The real identifier is gemini-3-pro-image, and it lives in our catalog as Nano Banana Pro.
This page is about whether half a minute and seven and a half cents is a good trade for your job. If you have already decided it is, the step-by-step calling guide is a separate post and it carries the code.
Where the thirty seconds goes
Pro composes interim "thought images" while it works. It roughs something in, looks at it, adjusts, and only then renders the frame you receive. Google does not return those intermediate images and we do not bill them. You pay for one output, and the model may well have made four on the way to it.
That reframes the latency. Pro is not badly optimised, it just spends its time in a different place: inside the request, instead of leaving the correction work to you afterwards. The alternative to 30 seconds of model deliberation is usually 30 minutes of yours. A flash-tier image that returns in two seconds with the headline spelled wrong has not saved you anything.
Budget for it honestly. A batch of 200 hero images is a coffee break, not a build step you block a deploy on.
The price, and what the price is against
Here is what a Nano Banana Pro request actually costs on E2X, at every resolution the model offers.
| Request | E2X | List price | Notes |
|---|---|---|---|
| 1K output | $0.075 | $0.15 | Base rate |
| 2K output | $0.075 | $0.15 | Same price as 1K |
| 4K output | $0.15 | $0.30 | Doubles, up to 4096×4096 |
| Interim thought images | $0.00 | — | Never billed, never returned |
Prices and product terms can change. Check each provider's current pricing before making a purchasing decision. Google Batch or Flex pricing is not directly equivalent to a standard on-demand API request because scheduling, availability, and processing conditions differ; it is therefore excluded from this comparison. This is a scoped comparison, not a claim that E2X is the world's cheapest option in every configuration.
Read the second row again. 1K and 2K cost exactly the same here, which is true nowhere else in our Google lineup — Nano Banana 2 charges a 1.5× multiplier for 2K. On Pro, 2K is free, and most integrations we see are still sending 1k because somebody copied a config from a flash-tier project. Send 2k. There is no reason not to.
The discount against list is real and we will take credit for it, but it is the wrong number to stare at. The competitor for a Pro request is rarely a cheaper model. It is a person opening a design tool.
Two capabilities, and what each one removes
Everything above is context. These are the two things you are actually buying, and if neither applies to your work then Pro is a bad purchase and we will say so in the next section.
Legible, styled, multilingual text inside the image. This is the headline capability and it is genuinely hard. Feed it an English infographic, ask for the same thing in Spanish, and what comes back is the original artwork with different words in it. The layout holds. Type lands where type belongs, and it reads as language rather than letterform soup. Localising a campaign into eight markets used to mean eight trips through a design tool with a translator's spreadsheet open. It is now eight requests and sixty cents. Compare that against the two days, not against a cheaper API.
The flash tiers render in-image text too, since the capability landed across the whole Gemini 3 generation. The 2.5-era Nano Banana has no version of it, having shipped before the trick existed. What Pro adds is reliability at length and across scripts. A single word survives on a flash tier. A paragraph in Japanese usually does not.
Fourteen reference images, sorted into named jobs. Other tiers take references as an undifferentiated pile. Pro spends a separate allowance on each kind of work: five slots for faces it has to keep recognisable, six for objects it has to reproduce exactly, three for imagery that only sets the look. It knows which pile is which.
That is the difference between "here are some pictures, good luck" and an art direction brief. A specific person, a specific bottle and a specific photographic treatment go into one request and come back as a composite that respects all three. One request. Not a generation, then a manual composite, then a colour pass.
If you are chasing the same face across a whole asset set, that character slot is the mechanism, and we go deeper on it in the consistent character generation guide.
The purchase we would talk you out of
Run this test before you integrate Pro. Does the finished image contain text a human needs to read? Does it need to combine references that play different roles — a person, a product, a treatment?
If both answers are no, do not buy this tier. You are paying a premium for two capabilities you will not call.
The alternatives, priced on 26 August 2026:
- Nano Banana 2 at $0.04 — up to 4K, fourteen references without the role split, and it renders in-image text well enough for short strings. This is the default for most production image work and it is a bit over half the price of Pro.
- Nano Banana 2 Lite at $0.0238 — 1K only, no
resolutionparameter at all, but it carries four extreme aspect ratios (1:4, 4:1, 1:8, 8:1) that no other model we run will produce. For banners, vertical rails and high-volume thumbnails, nothing on our catalog beats it on cost. - GPT Image 2 at $0.0525 — worth a look only for one specific reason, which is transparent PNG output. None of the Google tiers expose a
backgroundparameter. If you need a cut-out product shot or a sticker sheet in one call, this is the model. - Nano Banana Pro's edit endpoint at $0.075 — same price as generation, so if you are already on Pro there is no cost argument for editing on a cheaper tier and then upscaling.
The arithmetic gets loud at volume. Ten thousand images costs $750 on Pro, $400 on Nano Banana 2, and $238 on Lite. If you are generating ten thousand product thumbnails with no text in them, the difference between the top and bottom of that list is $512 you spent on deliberation you did not need.
We would rather tell you that now than watch you find out on the invoice. For the tier-by-tier breakdown across all four models, we wrote a full comparison.
What a request looks like
Enough shape to judge the integration cost, no more. Submit a job, get an ID back, then poll or take a webhook.
curl -X POST https://api.e2x.ai/v1/jobs/submit \
-H "Authorization: Bearer $E2X_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "google/nano-banana-pro/text-to-image",
"input": {
"prompt": "A ceramic espresso cup on a marble counter, morning light",
"aspect_ratio": "16:9",
"resolution": "2k"
}
}'
Two things in there will bite you if you skip them. The aspect_ratio default on this model is 9:16, not 1:1, so an unpinned config quietly produces portrait forever. And resolution takes lowercase values — 1k, 2k, 4k.
That is deliberately the only code here. The complete calling tutorial covers response parsing, polling, webhooks, the edit variant, every parameter and its default, and the storage gotcha that breaks pipelines a week after launch. The machine-readable spec is there for coding agents.
One thing to settle before you integrate
Google stamps SynthID into the pixels of everything its image models make. No flag switches it off — not at us, not at any reseller, not at Google's own endpoint. Most of the time this never comes up. It comes up hard when a client contract says deliverables ship unmarked, so settle that clause before you integrate rather than at handover. The rule holds across every Google model in our text-to-image category. The full catalog is here.
So, is it fast and affordable
Not fast. About 30 seconds, by design, and you would not want the version that skips the thinking.
Affordable is a yes with a condition. At $0.075 it is the most expensive image model we sell, roughly double Nano Banana 2 and three times Lite. Against those it is a luxury. Against a designer localising an infographic into eight languages by hand, it is not close. Pick the comparison that matches your actual alternative.
Frequently asked questions
How much does the Nano Banana Pro API cost?
E2X charges $0.075 per request for Nano Banana Pro at 1K or 2K resolution, against a $0.15 list price, checked 26 August 2026. A 4K request doubles to $0.15. Interim "thought images" the model generates while composing are never returned and never billed, so one request is one charge regardless of how much internal work it did.
Is Nano Banana Pro fast?
No, and that is intentional. A Pro request takes roughly 30 seconds end to end on our pipeline, compared with two to three seconds for the flash-tier Nano Banana models. The extra time is the model composing interim thought images before it renders the final frame. If raw speed matters more than the output, Nano Banana 2 Lite at $0.0238 is the faster and cheaper choice.
What is the model ID for Nano Banana Pro?
The Google model identifier is gemini-3-pro-image. It is often written as "Gemini 3.0 Pro", which is not a model Google ships. On E2X the slugs are google/nano-banana-pro/text-to-image for generation and google/nano-banana-pro/edit-image for editing.
Is Nano Banana Pro the same as Nano Banana 2?
No. They are different models on different weights with different prices — Pro is gemini-3-pro-image at $0.075, Nano Banana 2 is gemini-3.1-flash-image at $0.04. Some catalogs list Pro under the "Nano Banana 2" label and it is simply a mislabel. Pro's distinguishing feature is the role-separated reference budget, not the version number.
How many reference images does Nano Banana Pro accept?
Fourteen, and they are not interchangeable. Five slots are reserved for faces the model has to keep recognisable, six for objects it has to reproduce exactly, and three for imagery that only defines the look. Every other tier pools its fourteen into one flat set. That separation is what lets one request hold a named person, a named product and a named treatment at once.
Should I send 1K or 2K to Nano Banana Pro?
Send 2K. Pro is the only tier in our Google lineup where that step up is free: both settings land on $0.075, where Nano Banana 2 adds a 1.5× multiplier for the same jump. Asking for 1k here means full price for a quarter of the pixels. The values are lowercase: 1k, 2k, 4k.
When is Nano Banana Pro not worth the money?
When the image contains no text a reader needs and no multi-reference composite. Those two capabilities are what the premium pays for, and if neither is in play you are buying deliberation you will not use. Nano Banana 2 at $0.04 handles most production image work, and Nano Banana 2 Lite at $0.0238 handles high-volume 1K output. On ten thousand images that difference is $350 to $512.