Image Generation and Edits

Image endpoints require an OpenAI group with Images API support. Call /v1/models first to verify that the current key can see an image model.

Generate an image

curl https://codexapi.asia/v1/images/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "A clean studio photograph of a white ceramic cup.",
    "size": "1024x1024",
    "quality": "high",
    "response_format": "b64_json"
  }'

The generated image is usually returned in data[0].b64_json. Some chat clients display only text and do not render Base64 image data automatically.

Decode with Python

import base64
import os

from openai import OpenAI

client = OpenAI(
    api_key=os.environ["CODEX_RELAY_API_KEY"],
    base_url="https://codexapi.asia/v1",
)

result = client.images.generate(
    model="gpt-image-2",
    prompt="A minimal product photo of a desk lamp.",
    size="1024x1024",
    response_format="b64_json",
)

with open("image.png", "wb") as image_file:
    image_file.write(base64.b64decode(result.data[0].b64_json))

Edit an image

Image edits use a multipart form:

curl https://codexapi.asia/v1/images/edits \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "model=gpt-image-2" \
  -F "prompt=Change the background to pure white" \
  -F "image=@input.png" \
  -F "response_format=b64_json"

Responses image tool

Models with Responses tool support can accept an image_generation tool in a /v1/responses request. Availability depends on the model, group, and upstream account capabilities.

The request succeeded but no image is visible

1. Inspect the raw JSON for b64_json or an image-tool result.
2. Confirm that the client can render Base64 images.
3. Save the image with the Python example to isolate a display issue.
4. If the response contains no image field, keep the request ID and verify that the selected model supports image output.