Nano Banana API — use the OpenAI SDK as-is, per-image pricing and calling method

Search for the Nano Banana API in Korean and the results split in two: Google's official documentation or blog posts about “how to get a key in AI Studio.” What you actually need lies between those two. This article lays out how to make calls while keeping your existing OpenAI SDK and lists the per-image prices across the full model family in one table. No Google Cloud project is required.

Per-image prices — the full model family

The rates below are read from the model catalog, not hard-coded into the article. The table updates when prices change.

All prices are based on per image and are deducted from prepaid balance, not a subscription. There is no charge in months when you generate nothing.

Make a request — change only base_url and the model name

If you already use the OpenAI SDK, there is nothing new to learn. Just change two things in the client initialization: base_url and the model name.

# 기존 OpenAI SDK를 그대로 씁니다. base_url과 모델 이름만 바뀝니다.
# Google Cloud 프로젝트도, 별도 SDK 설치도 필요 없습니다.
import os
from openai import OpenAI

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

img = client.images.generate(
    model="nano-banana-2",
    prompt="한옥 마당에 눈이 내리는 장면, 따뜻한 조명, 사진 스타일",
    n=1,
)
print(img.data[0].url)

The endpoint is POST /v1/images/generations, in the same format as OpenAI's image endpoint. So if you already have image generation integrated, you can switch by changing just the model name string.

First, check that your key works

Before changing your code, verify your key and endpoint with one request. Then, if an error occurs later, you can narrow down the cause to the code.

# 키가 되는지 30초 만에 확인하는 최소 요청
curl -sS https://api.kunavo.com/v1/images/generations \
  -H "Authorization: Bearer sk-kn-..." \
  -H "content-type: application/json" \
  -d '{"model":"nano-banana","prompt":"a paper crane on a desk","n":1}'

A 401 means an authentication issue, 429 means rate limiting, and 529 means upstream overload. The decision table in Claude Code Error Guide applies here as well.

Which tier should you use by default?

The most cost-effective approach is to use the base model by default and upgrade to Pro only for jobs where the results are insufficient. Since you only need to change one model name string, comparison testing costs virtually nothing — run the same prompt once on each tier and decide.

Video pricing is covered in Veo 3 Pricing, and the detailed English documentation is in Nano Banana API. Payment methods and registering domestic cards are covered in Claude API Pricing and Payments.

Frequently asked questions