Image generation

Call GPT Image, Gemini (Nano Banana) and other image models with the OpenAI Images format, at 1K / 2K / 4K.

Endpoint

POST https://api.onefast.ai/v1/images/generations   # text-to-image
POST https://api.onefast.ai/v1/images/edits         # image-to-image / inpainting (multipart)
Authorization: Bearer sk-xxx

All image models are called with the OpenAI Images format, and images are returned as Base64 (b64_json).

Text-to-image

cURL

curl https://api.onefast.ai/v1/images/generations \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2.5",
    "prompt": "A cute corgi astronaut, studio lighting",
    "size": "1024x1024",
    "n": 1
  }' | python3 -c "import sys,json,base64; open('out.png','wb').write(base64.b64decode(json.load(sys.stdin)['data'][0]['b64_json']))"

Python

import base64
from openai import OpenAI

client = OpenAI(api_key="sk-xxx", base_url="https://api.onefast.ai/v1")
result = client.images.generate(
    model="gpt-image-2.5",
    prompt="A cute corgi astronaut, studio lighting",
    size="1024x1024",
)
with open("out.png", "wb") as f:
    f.write(base64.b64decode(result.data[0].b64_json))

Node.js

import fs from "node:fs";
import OpenAI from "openai";

const client = new OpenAI({ apiKey: "sk-xxx", baseURL: "https://api.onefast.ai/v1" });
const result = await client.images.generate({
  model: "gpt-image-2.5",
  prompt: "A cute corgi astronaut, studio lighting",
  size: "1024x1024",
});
fs.writeFileSync("out.png", Buffer.from(result.data[0].b64_json, "base64"));

Image generation usually takes 20–60 seconds. Set your HTTP client timeout to 300 seconds or more.

Common parameters

Gemini images (Nano Banana) and 4K

Gemini (Nano Banana) image models such as gemini-3.1-flash-image, gemini-3-pro-image-preview and gemini-nano-banana-2.1 use the same endpoint. The gateway decides the resolution tier and aspect ratio as follows:

Resolution (1K / 2K / 4K) — quality takes precedence; otherwise the longest side of size decides:

Aspect ratio — taken from the width-to-height ratio of size and snapped to the nearest standard ratio (1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 21:9).

cURL

curl https://api.onefast.ai/v1/images/generations \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-3.1-flash-image",
    "prompt": "A watercolor lighthouse at dawn",
    "size": "3840x2160"
  }'

The example above produces a 16:9 4K image. Exact pixel dimensions are determined by the model (for example, 2K at 16:9 is about 2752×1536). The output may be PNG or JPEG — check the file header.

Image-to-image / editing

Upload the source image (multipart form) and describe the change in prompt:

cURL

curl https://api.onefast.ai/v1/images/edits \
  -H "Authorization: Bearer $API_KEY" \
  -F model="gpt-image-2.5" \
  -F image="@input.png" \
  -F prompt="Add a red scarf to the dog"

Python

result = client.images.edit(
    model="gpt-image-2.5",
    image=open("input.png", "rb"),
    prompt="Add a red scarf to the dog",
)

Gemini image models accept several reference images in one request (up to 14, field name image[]) and do not support mask.

Response and token usage

A successful response returns the images in data. GPT Image and Gemini image models also return token usage in usage:

{
  "created": 1793777279,
  "data": [{ "b64_json": "iVBORw0KGgo..." }],
  "usage": {
    "total_tokens": 1419,
    "input_tokens": 7,
    "output_tokens": 1412,
    "input_tokens_details": { "text_tokens": 7, "image_tokens": 0 },
    "output_tokens_details": { "text_tokens": 292, "image_tokens": 1120 }
  }
}

• For Gemini image models, usage covers the whole request, including every image when n is greater than 1. output_tokens includes thinking tokens; these, along with any output the model does not break down by type, are counted in text_tokens.
• input_tokens_details and output_tokens_details appear only when the model reports a breakdown. usage is omitted when no token usage is reported.
• For Gemini image models in groups that charge per token, the charge is based on the full input_tokens and output_tokens; input_tokens_details and output_tokens_details are only a breakdown.
• In groups that charge per image, usage is for reference only and does not affect the charge.

Billing

• Some groups charge per image (regardless of size), others per token (output image tokens at the output price). The listed price for that group in the model catalog applies.
• Failed requests are not billed.