Nano Banana API — call Google's image models OpenAI-compatibly at half the price

Nano Banana is Google's image-generation model family — the model behind Gemini's native image output. Kunavo serves the whole family (nano-banana, nano-banana-2, nano-banana-pro) through the OpenAI-compatible images endpoint, at roughly half of Google's list price, on one bearer token and one Stripe balance. If your code can call the OpenAI images API, it can call Nano Banana by changing only the base_url.

Call the Nano Banana API

Generate a key at /app/keys, keep the OpenAI SDK, and point base_url at Kunavo. Text-to-image is POST /v1/images/generations:

import os
from openai import OpenAI

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

img = client.images.generate(
    model="nano-banana",
    prompt="A neon ramen stall in the rain, cinematic, 35mm",
    size="1024x1024",
)
print(img.data[0].url)  # temporary URL (~24h) — download and re-host

Returned URLs are temporary (~24h) — download and re-host on your own CDN. Full parameters (size, resolution, quality, async + webhooks) are in the images API reference.

What Nano Banana costs

Nano Banana is billed per image, not per month — $0.0273 an image on Kunavo for the base model, against Google's ~$0.039 list. There is no subscription on the API route: you pay for images you generate and an idle month costs nothing. That is the first thing to settle, because most "Nano Banana pricing" results are describing consumer apps that wrap the model in a monthly plan with a credit allowance — a different product with a different bill.

Per-image rates below are about half of Google's official list. Live rates are on each model page.

At $0.0273 an image, 100 images cost about $2.73, 1,000 about $27.3 and 10,000 about $273 — the same arithmetic all the way up, because there is no tier to cross and no minimum to commit to. Kunavo bills from a prepaid balance with a $10 minimum top-up and no expiry, so the only way to spend more than you planned is to generate more images than you planned.

Two things change the number in practice. Resolution is a real price axis on nano-banana-2 and nano-banana-pro — the 1K / 2K / 4K tiers price separately, and the table above shows the entry tier. And a failed generation that never returns an image is not billed, but a generation you dislike is: prompt iteration is the cost most people forget, which is why drafting on nano-banana and re-rendering the winners on Pro is the cheapest workflow.

Nano Banana Pro API

Nano Banana Pro is the premium tier — the pick when output quality is the product: hero images, posters, ads, and anything with text rendered inside the image, where Pro is clearly ahead of the base models. The call is identical to the example above with model="nano-banana-pro"; resolution tiers (1K/2K/4K) price separately, from about $0.067+ per image against Google's ~$0.134 list — live rates on the nano-banana-pro model page, and the per-resolution breakdown plus what actually moves an image bill in Nano Banana Pro pricing. For volume drafts, generate with nano-banana first and re-render the winners with Pro; the two share prompts cleanly.

Edit an image (image-to-image)

POST /v1/images/edits takes a prompt plus a source image — an https URL or a data: base64 URI, and returns the edited result. Use nano-banana-edit for Nano Banana image-to-image.

# Image-to-image: pass a source image URL (or a data: base64 URI)
curl https://api.kunavo.com/v1/images/edits \
  -H "Authorization: Bearer $KUNAVO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nano-banana-edit",
    "prompt": "make the sky golden hour, keep the subject",
    "image": "https://example.com/source.png"
  }'

Kunavo vs calling Google directly

Nano Banana is Google's Gemini image model, also reachable first-party through the Gemini API. Same model, different wrapper:

The practical reasons teams land here: the OpenAI-shaped interface, no Google Cloud project, and the same key reaching GPT-Image-2, Veo 3 and Suno for A/B tests.

When to use which

• nano-banana — highest volume at the lowest cost; thumbnails, variations, drafts.
• nano-banana-2 — the default when fidelity matters and you want resolution tiers.
• nano-banana-pro — hero images, posters and anything with text baked into the image.
• For the full multimodal picture — image → video → voiceover pipelines — see the multimodal AI guide.

FAQ