Skip to main content

​
Quick Overview

gpt-image-2 is an image generation and image editing model available through LaoZhang API. Use it when you need an OpenAI-compatible path that can cover both text-to-image and image-to-image workflows.

Price

gpt-image-2 is billed per request at $0.03/call.

Online Test

Test the model online at yingtu.ai before integrating the API.

​
Which Endpoint Should You Use?

NeedRecommended endpointResponse shape
You already use Chat Completions and want minimal changes/v1/chat/completionsMarkdown image URL in choices[0].message.content
You only need text-to-image with the standard Images API shape/v1/images/generationsdata[0].b64_json by default, or data[0].url
You need multipart local image upload for image editing/v1/images/editsdata[0].b64_json by default, or data[0].url
Your source image is already a CDN URL or base64 data URL/v1/chat/completionsMarkdown image URL
If your app only needs “prompt in, image URL out”, start with /v1/chat/completions. If you already have OpenAI Images API code, use /v1/images/generations and /v1/images/edits.

​
Base Configuration

export LAOZHANG_API_KEY="sk-your-key"
export BASE_URL="https://api.laozhang.ai/v1"
Use the unified OpenAI-compatible gateway URL https://api.laozhang.ai/v1. Do not set the base URL to a model-level path.

​
Chat Completions Text-to-Image

This is the simplest integration path. Set model to gpt-image-2 and put the image request in messages. 
curl "$BASE_URL/chat/completions" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $LAOZHANG_API_KEY" \
  -d '{
    "model": "gpt-image-2",
    "messages": [
      {
        "role": "user",
        "content": "Create a clean product render of a white ceramic mug on a gray desk, soft natural light, minimal background"
      }
    ],
    "stream": false
  }'
The generated image usually appears as a Markdown image link in choices[0].message.content: 
{
  "choices": [
    {
      "message": {
        "role": "assistant",
        "content": "![image](https://example.com/generated.png)"
      }
    }
  ]
}

​
Chat Completions Image-to-Image

Use /v1/chat/completions when your source image is available as a CDN URL or a base64 data URL. In this mode, messages[].content is an array that includes both the text instruction and the image input.

​
Use an Image URL

curl "$BASE_URL/chat/completions" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $LAOZHANG_API_KEY" \
  -d '{
    "model": "gpt-image-2",
    "messages": [
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "Use the provided image as the source. Keep the subject, composition, and text. Only change the sticker to red and add a very thin gold rim to the mug."
          },
          {
            "type": "image_url",
            "image_url": {
              "url": "https://example.com/source.png"
            }
          }
        ]
      }
    ],
    "stream": false
  }'

​
Use a base64 Data URL

curl "$BASE_URL/chat/completions" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $LAOZHANG_API_KEY" \
  -d '{
    "model": "gpt-image-2",
    "messages": [
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "Use the provided image as the source. Keep the subject, composition, and text. Only change the sticker to green."
          },
          {
            "type": "image_url",
            "image_url": {
              "url": "data:image/png;base64,put-image-base64-here"
            }
          }
        ]
      }
    ],
    "stream": false
  }'

​
Images Generations Text-to-Image

If your code already uses the OpenAI Images API shape, call /v1/images/generations. The default response returns data[0].b64_json: 
curl "$BASE_URL/images/generations" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $LAOZHANG_API_KEY" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "Create a clean product poster: a white card on a wooden desk with IMG-42 printed in the lower-right corner"
  }'
Add response_format when you want a URL: 
curl "$BASE_URL/images/generations" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $LAOZHANG_API_KEY" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "Create a clean product poster: a white card on a wooden desk with IMG-42 printed in the lower-right corner",
    "response_format": "url"
  }'

​
Images Edits Image-to-Image

Use /v1/images/edits when your source image is a local file and multipart upload is the most convenient path. The default response returns data[0].b64_json: 
curl "$BASE_URL/images/edits" \
  -H "Authorization: Bearer $LAOZHANG_API_KEY" \
  -F "model=gpt-image-2" \
  -F "prompt=Use the provided image as the source. Keep the desk, card, and IMG-42 label. Only add a very thin red border to the card." \
  -F "image=@source.png"
Return a URL: 
curl "$BASE_URL/images/edits" \
  -H "Authorization: Bearer $LAOZHANG_API_KEY" \
  -F "model=gpt-image-2" \
  -F "prompt=Use the provided image as the source. Keep the desk, card, and IMG-42 label. Only add a very thin blue border to the card." \
  -F "image=@source.png" \
  -F "response_format=url"

​
Parse Responses

​
Extract a Chat Completions Image URL

import re

content = response["choices"][0]["message"]["content"]
match = re.search(r"!\[[^\]]*\]\((https?://[^)\s]+)\)", content)
image_url = match.group(1) if match else None

​
Save Images API b64_json

import base64

value = response["data"][0]["b64_json"]
if value.startswith("data:"):
    value = value.split(",", 1)[1]

with open("output.png", "wb") as f:
    f.write(base64.b64decode(value))

​
Read an Images API URL

image_url = response["data"][0]["url"]

​
Prompt Pattern for Image Editing

For image-to-image, tell the model what to preserve and what to change. A reliable pattern is: 
Use the provided image as the source. Keep the subject, composition, camera angle, and key text. Change only one specific detail: turn X into Y.
More stable instructions:
  • Keep the product, background, and camera angle; change only a color or local decoration.
  • Keep the card position and text; add only a border, label, or small icon.
  • Avoid “generate a similar image”, which can encourage the model to redraw the whole image.

​
FAQ

/v1/chat/completions returns a chat completion shape. The image URL is inside the Markdown image link at choices[0].message.content. data[0].url belongs to Images API responses.
Check whether the value includes a data:image/png;base64, prefix. If it does, split on the first comma and decode only the second part.
gpt-image-2 is billed per request at $0.03/call.
Yes. Visit yingtu.ai to test prompts online, then move the confirmed prompt into your API call.