GPT Image 2

GPT Image 2 is OpenAI's latest GPT Image generation and editing model. CiYuanX exposes it through an OpenAI-compatible image API, so with the official OpenAI SDK you only need to point base_url / baseURL at https://ciyuanx.io/v1.

Capability	Endpoint
Generate images	POST https://ciyuanx.io/v1/images/generations
Edit images	POST https://ciyuanx.io/v1/images/edits

Model ID: gpt-image-2

Choosing an API

If you only need to generate or edit an image in a single request, the Image API is all you need — and it's the recommended path for most CiYuanX integrations. OpenAI also offers an image-generation tool in the Responses API for conversational, multi-step editing workflows. If your product needs multi-turn editing, first confirm whether the corresponding model and endpoint support the Responses API on CiYuanX.

Generate images

The Image API returns base64 image data in data[0].b64_json. Decode it and write it to a file.

Python

from openai import OpenAI
import base64

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://ciyuanx.io/v1",
)

result = client.images.generate(
    model="gpt-image-2",
    prompt="A clean product photo of a matte black smart desk lamp on a white background",
    size="1024x1024",
    quality="medium",
)

image_bytes = base64.b64decode(result.data[0].b64_json)
with open("lamp.png", "wb") as f:
    f.write(image_bytes)

TypeScript

import OpenAI from "openai";
import { writeFileSync } from "node:fs";

const client = new OpenAI({
  apiKey: "YOUR_API_KEY",
  baseURL: "https://ciyuanx.io/v1",
});

const result = await client.images.generate({
  model: "gpt-image-2",
  prompt: "A clean product photo of a matte black smart desk lamp on a white background",
  size: "1024x1024",
  quality: "medium",
});

writeFileSync("lamp.png", Buffer.from(result.data[0].b64_json, "base64"));

cURL

curl https://ciyuanx.io/v1/images/generations \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "A clean product photo of a matte black smart desk lamp on a white background",
    "size": "1024x1024",
    "quality": "medium"
  }'

The image below was produced by the Python example above and saved by decoding data[0].b64_json:

Edit with reference images

Pass one or more images to images.edit and describe the result you want in text. Multiple input images can serve as references for a new composition.

Python

from openai import OpenAI
import base64

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://ciyuanx.io/v1",
)

result = client.images.edit(
    model="gpt-image-2",
    image=[
        open("product.png", "rb"),
        open("background.png", "rb"),
    ],
    prompt="Place the product from the first image onto the desk in the second image, keeping the lighting realistic.",
    size="1536x1024",
    quality="high",
)

image_bytes = base64.b64decode(result.data[0].b64_json)
with open("edited-product.png", "wb") as f:
    f.write(image_bytes)

TypeScript

import OpenAI from "openai";
import { createReadStream, writeFileSync } from "node:fs";

const client = new OpenAI({
  apiKey: "YOUR_API_KEY",
  baseURL: "https://ciyuanx.io/v1",
});

const result = await client.images.edit({
  model: "gpt-image-2",
  image: [
    createReadStream("product.png"),
    createReadStream("background.png"),
  ],
  prompt: "Place the product from the first image onto the desk in the second image, keeping the lighting realistic.",
  size: "1536x1024",
  quality: "high",
});

writeFileSync("edited-product.png", Buffer.from(result.data[0].b64_json, "base64"));

cURL

curl https://ciyuanx.io/v1/images/edits \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F model="gpt-image-2" \
  -F "image[]=@product.png" \
  -F "image[]=@background.png" \
  -F prompt="Place the product from the first image onto the desk in the second image, keeping the lighting realistic." \
  -F size="1536x1024" \
  -F quality="high"

The example below composites a product shot with a desk background and returns a new 1536×1024 PNG:

Product input	Background input	Edited output

Edit with a mask

When you want to guide the model to change only a specific region, pass a mask. The source image and the mask must have the same size and format, and the mask must include an alpha channel.

Python

from openai import OpenAI
import base64

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://ciyuanx.io/v1",
)

result = client.images.edit(
    model="gpt-image-2",
    image=open("room.png", "rb"),
    mask=open("mask.png", "rb"),
    prompt="Replace the masked area with a modern walnut bookshelf.",
)

image_bytes = base64.b64decode(result.data[0].b64_json)
with open("room-bookshelf.png", "wb") as f:
    f.write(image_bytes)

cURL

curl https://ciyuanx.io/v1/images/edits \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F model="gpt-image-2" \
  -F image=@room.png \
  -F mask=@mask.png \
  -F prompt="Replace the masked area with a modern walnut bookshelf."

The mask uses its alpha channel to mark the editable region; the endpoint returns a regular PNG:

Source	Mask	Edited output

Streaming partial images

OpenAI's upstream API offers stream and partial_images for receiving intermediate images before the final one is ready. On CiYuanX this is a compatibility-dependent capability: unless you've confirmed that the endpoint already supports streamed image events, prefer a standard non-streaming call.

Even if you pass stream: true and partial_images, the response may still be a regular final JSON payload that returns the image via data[0].b64_json.

Output options

Parameter	Values	Notes
size	auto, 1024x1024, 1536x1024, 1024x1536, 2048x2048, 2048x1152, 3840x2160, 2160x3840, or another custom size that meets the constraints	Each side must be a multiple of 16 px, the longest side at most 3840 px, the aspect ratio no more than 3:1, and the total pixel count between 655,360 and 8,294,400.
quality	auto, low, medium, high	Use low for drafts and fast iteration; prefer medium or high for final assets.
output_format	png, jpeg, webp	Defaults to png. When you don't need a transparent background, jpeg is usually faster.
output_compression	0 to 100	Applies only to jpeg and webp.
background	auto, opaque background options	gpt-image-2 does not currently support background: "transparent".
stream, partial_images	Compatibility-dependent	CiYuanX may return a regular final JSON payload instead of streamed partial-image events.
moderation	auto, low	auto is the default filtering level.

Notes and limitations

• Complex prompts can take longer to process; set a reasonable timeout or use a background job on your side.
• The model's text rendering has improved but still isn't guaranteed to be accurate. Review logos, labels, and UI mockups by hand.
• Characters, exact layout, and brand details may not stay perfectly consistent across multiple generations.
• Don't pass input_fidelity when editing with gpt-image-2; the model handles input images at high fidelity automatically.
• A mask guides the edit region — it isn't a pixel-perfect selection.