Nano Banana Pro API: Google's premium image model, one endpoint.

Quick answer: what Pro is and when it earns the call.

Nano Banana Pro is the premium member of Google's Nano Banana family, the marketing name for Gemini's native image models. Google positions it explicitly as the choice for the most complex visual tasks: the highest level of world knowledge, advanced localization, accurate brand consistency, and precision creative control. It shipped in late 2025 as Gemini 3 Pro Image and, since the February 2026 release of Nano Banana 2, sits above the newer generalist as the quality ceiling of the family. The one-line routing rule: nano-banana-2 by default, nano-banana-pro when the output is the product. Marketing assets that carry a brand, infographics where every label must be right, localized visuals where translated text has to render cleanly, compositions with many constrained elements. For drafts, volume, and everyday illustration, the generalist is faster and costs less per image; Pro is the model you promote a job to, not the one you start every job on.

What Pro does that the base tier does not, verified.

Two hedges worth stating plainly. First, Pro is slower than the base tier; Google trades latency for quality at this tier, and complex prompts compound it, so treat Pro as a seconds-scale model in your UX budget rather than assuming any specific number. Second, capability claims here are Google's positioning verified against its documentation as of July 2026; for your workload, the only benchmark that matters is the same prompt run on both ids, which the shared endpoint makes a two-line experiment.

• Text rendering at localization grade: Google names advanced text rendering and localization as Pro's headline strengths. Where nano-banana-2 renders reliable short strings, Pro is the model Google points at dense, multilingual, layout-sensitive text: infographics, posters, packaging.
• Brand and character consistency: Pro is documented for accurate brand consistency and supports reference-image inputs, up to 14 reference images per Google's API docs, for holding a subject or identity stable across generations.
• World knowledge and reasoning: as a Gemini 3 Pro derivative, Pro brings the strongest factual grounding of the family, which shows up in images that depict real things, maps, diagrams, historical detail, plausibly.
• Precision control: Google positions Pro for deliberate composition, where element placement, lighting direction, and camera geometry are specified rather than suggested.
• Resolution: 1K, 2K, and 4K output tiers; the 512px efficiency tier is a Nano Banana 2 feature, not a Pro one.

The calling pattern.

Pro speaks the standard OpenAI image-generation shape through the gateway: model nano-banana-pro, a prompt, and a size. The response returns the image in data[0] as a base64 payload or URL. Editing goes through the images edit route with an input image attached. Because Pro is the expensive tier, the pattern worth copying is the draft-then-promote loop below: iterate on the concept with nano-banana-2, and re-render the approved prompt on nano-banana-pro at final resolution. The prompt carries over verbatim; only the model string and size change.

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["APISROUTER_API_KEY"],
    base_url="https://api.apisrouter.com/v1",
)

prompt = (
    "Conference poster, headline 'SHIP WEEK 2026' in bold condensed type, "
    "three labeled schedule blocks, Swiss grid layout, off-white paper, "
    "single accent color"
)

# iterate on the concept at the generalist tier
draft = client.images.generate(model="nano-banana-2", prompt=prompt, size="1024x1024")

# promote the approved prompt to Pro at final resolution
final = client.images.generate(model="nano-banana-pro", prompt=prompt, size="2048x2048")

Prompting Pro: spend the control it gives you.

Pro rewards specificity more than any other model in the family. Where the base tier fills gaps with sensible defaults, Pro is built to follow deliberate direction, so give it some: name the layout system, the type feel, the lighting, the camera, the exact strings to render and where they sit. A Pro prompt reads less like a wish and more like a creative brief. For in-image text, the reliable pattern is quoting every string and stating its role: headline, label, caption. Pro's localization strength means translated text is a first-class request, "the same poster with the headline in Japanese", which is a genuinely hard case for most image models and the kind of job that justifies the tier. For brand work, feed reference images where the integration supports image inputs, and describe the brand constraints in words as well: exact color descriptions, logo placement, what must not change. Pro's consistency is a capability, not a guarantee; the constraint text is what it anchors to. Note that reference-image workflows are richest on Google's native API; through the stateless gateway surface, pass references as edit inputs and keep iterations explicit.

What Pro costs, and how to keep it rational.

Google's published Gemini API rates, checked July 2026, price Nano Banana Pro at $0.134 per image at 1K and 2K resolutions and $0.24 per image at 4K. Against the base tier, that is roughly double the per-image rate of nano-banana-2 at 1K, and the gap widens with volume. For what the same id costs through this gateway, the pricing page carries current per-image rates; image billing is per image, not per token, and draws from the same prepaid balance as your text traffic. Keeping Pro rational is a routing exercise. The draft-then-promote loop keeps iteration off the premium tier. Resolution discipline keeps 4K spend on final assets only. And per-key usage in the console makes the split visible: a key per pipeline shows exactly how many images ran on which tier, so the moment Pro traffic creeps into draft work, the log says so.

Use cases where Pro is the right first call.

The common thread is that in each case a human will inspect the image closely, or the image carries text or identity that must be exact. When neither holds, the base tier's speed and price win.

• Infographics and data visuals: many labeled elements, dense text, and factual content lean on Pro's text rendering and world knowledge at once.
• Brand asset generation: campaign images that must hold a logo, palette, and product identity stable across a series.
• Localized marketing visuals: the same creative re-rendered per language, with translated headlines that have to typeset cleanly.
• Print and 4K deliverables: posters, packaging mockups, and anything that ships at high resolution where artifacts are visible.
• Editorial composites: deliberate multi-element scenes, specified camera and lighting, where "close enough" is not.

Verify access and debug the first Pro call.

List /v1/models with your key and confirm nano-banana-pro appears; the catalog listing is the source of truth for the exact id your key can address. A model-not-found error on a first call is nearly always an id mismatch with that listing. Latency behaves differently at this tier: complex Pro generations run seconds to tens of seconds, so set client timeouts generously, sixty seconds is a sane floor, and do not retry on a timer shorter than the model's honest working time, or you will pay for images you abandoned. Parameter 400s follow the usual compatibility-layer rule: reduce to model, prompt, size, confirm, then add options back one at a time. And once traffic flows, the console's per-request log shows model, count, and spend, which for a per-image billed tier is the fastest sanity check that your routing policy is doing what you designed.

curl -s https://api.apisrouter.com/v1/models \
  -H "Authorization: Bearer $APISROUTER_API_KEY" | grep nano-banana

curl -s https://api.apisrouter.com/v1/images/generations \
  -H "Authorization: Bearer $APISROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"nano-banana-pro",
       "prompt":"Minimal infographic, three steps, labeled arrows",
       "size":"1024x1024"}' --max-time 90

FAQ