Qwen Image

Qwen Image 3.0 is Alibaba Cloud’s image model. It generates an image from a text prompt, and it edits up to three images that you send with the prompt. The call is synchronous: one POST returns the image in the response, so there is nothing to poll. The request and response shapes mirror Alibaba Cloud’s own API.

Best for

Generating and editing on one endpoint, with one model ID.
Edits that combine up to three input images.
Batches. One request returns up to six images.

Models

Model	Model ID	Tasks	Sizes	Input images	Images per request
Qwen Image 3.0	`qwen-image-3.0`	Generate, edit	`1K`, `2K`	Up to 3	1 to 6
Qwen Image 3.0 Pro	`qwen-image-3.0-pro`	Generate, edit	`1K`, `2K`	Up to 3	1 to 6

Each model uses one ID for both tasks. The items in the message select the task. A text item alone generates an image. Image items before the text item edit those images. Qwen Image 3.0 Pro is the higher grade of Qwen Image 3.0, with the same request shape and the same limits. It costs more per image. Use Qwen Image 3.0 while you iterate, and Qwen Image 3.0 Pro for the final image.

Endpoints

The base URL is https://api.nunchux.ai.

Step	Method	Path
Generate or edit	`POST`	`/v1/alibaba/services/aigc/multimodal-generation/generation`

There is no poll route. The image is in the response. Wan and HappyHorse use the same /v1/alibaba route family, with an asynchronous video route.

Request

Send your API key in the X-API-Key header. See Authentication. Set Content-Type: application/json.

string

required

The model ID from the table above: qwen-image-3.0 or qwen-image-3.0-pro.

object

required

The prompt and the input images.

Show properties

array

required

One message, with role set to user.

Show message properties

string

required

Must be user.

array

required

The input images first, then the prompt.

{"image": "..."}: an input image, as a URL or a data URI. Edits only. Up to 3 items. The provider fetches a URL, so it must be reachable from the internet.
{"text": "..."}: the prompt. One item, after the images. On an edit, say what must change and what must stay.

object

The generation controls.

Show properties

string

The output size, as width*height in pixels, for example 1328*1328. The output area selects the price tier. A size up to 2,250,000 pixels is billed at the 1K tier. A larger size is billed at the 2K tier. Alibaba Cloud documents sizes from 512*512 to 2048*2048.If you omit it, the image is 2048*2048 and is billed at the 2K tier.

integer

default:1

The number of images. Range: 1 to 6. Each image is billed.

boolean

default:true

When true, Alibaba Cloud rewrites the prompt into a more detailed one before the image renders, and output.rewrite_status is success. Set it to false to render your prompt as written. output.rewrite_status is then not_use.

Common sizes

Shape	`1K` tier	`2K` tier
1:1	`1328*1328`	`2048*2048`
16:9	`1664*928`	`2048*1152`
9:16	`928*1664`	`1152*2048`
4:3	`1472*1104`	`2048*1536`
3:4	`1104*1472`	`1536*2048`

A minimal request body:

{
  "model": "qwen-image-3.0",
  "input": {
    "messages": [
      {
        "role": "user",
        "content": [{ "text": "a paper lantern floating over a quiet canal at night" }]
      }
    ]
  },
  "parameters": { "size": "1328*1328", "n": 1 }
}

Response

The response is Alibaba Cloud’s own JSON body. Each image is a URL in output.choices[].message.content[].

object

required

The result.

Show properties

array

required

The generated images.

Show item properties

string

The URL of the image. It is signed and expires after 24 hours. Download the image as soon as you receive it.

string

success when Alibaba Cloud rewrote the prompt, not_use when prompt_extend was false.

object

What the request produced: output_width, output_height, output_image_count, output_image_type (the size tier), and input_image_count. Nunchux bills from these values.

string

required

Alibaba Cloud’s identifier for the request.

{
  "output": {
    "choices": [
      {
        "finish_reason": "stop",
        "message": {
          "role": "assistant",
          "content": [{ "type": "image", "image": "https://<signed-url>" }]
        }
      }
    ],
    "rewrite_status": "success"
  },
  "usage": {
    "input_image_count": 0,
    "input_image_type": "qima_input_1k",
    "output_height": 1328,
    "output_image_count": 1,
    "output_image_type": "qima_output_1k",
    "output_width": 1328
  },
  "request_id": "19a65aa2-7483-9dc9-85ab-92f349127856"
}

Example

Generate

# Synchronous: one POST returns the image URL. There is no polling.
# The URL is at output.choices[0].message.content[0].image and expires after 24 hours.
url=$(curl -s https://api.nunchux.ai/v1/alibaba/services/aigc/multimodal-generation/generation \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $NUNCHUX_API_KEY" \
  -d '{
    "model": "qwen-image-3.0",
    "input": {
      "messages": [
        {
          "role": "user",
          "content": [{"text": "a paper lantern floating over a quiet canal at night"}]
        }
      ]
    },
    "parameters": {"size": "1328*1328", "n": 1}
  }' | jq -r '.output.choices[0].message.content[0].image')
curl -s -o output.png "$url"

import os
import requests

response = requests.post(
    "https://api.nunchux.ai/v1/alibaba/services/aigc/multimodal-generation/generation",
    headers={
        "Content-Type": "application/json",
        "X-API-Key": os.environ["NUNCHUX_API_KEY"],
    },
    json={
        "model": "qwen-image-3.0",
        "input": {
            "messages": [
                {
                    "role": "user",
                    "content": [{"text": "a paper lantern floating over a quiet canal at night"}],
                }
            ]
        },
        "parameters": {"size": "1328*1328", "n": 1},
    },
)

# The URL expires after 24 hours, so download the image at once.
url = response.json()["output"]["choices"][0]["message"]["content"][0]["image"]
with open("output.png", "wb") as f:
    f.write(requests.get(url).content)

import { writeFile } from "node:fs/promises";

const response = await fetch(
  "https://api.nunchux.ai/v1/alibaba/services/aigc/multimodal-generation/generation",
  {
    method: "POST",
    headers: {
      "Content-Type": "application/json",
      "X-API-Key": process.env.NUNCHUX_API_KEY,
    },
    body: JSON.stringify({
      model: "qwen-image-3.0",
      input: {
        messages: [
          {
            role: "user",
            content: [{ text: "a paper lantern floating over a quiet canal at night" }],
          },
        ],
      },
      parameters: { size: "1328*1328", n: 1 },
    }),
  }
);
const data = await response.json();

// The URL expires after 24 hours, so download the image at once.
const url = data.output.choices[0].message.content[0].image;
await writeFile("output.png", Buffer.from(await (await fetch(url)).arrayBuffer()));

Edit

Put the input images before the prompt. Replace the example URL with an image that the provider can fetch.

url=$(curl -s https://api.nunchux.ai/v1/alibaba/services/aigc/multimodal-generation/generation \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $NUNCHUX_API_KEY" \
  -d '{
    "model": "qwen-image-3.0-pro",
    "input": {
      "messages": [
        {
          "role": "user",
          "content": [
            {"image": "https://example.com/street.png"},
            {"text": "make it night and add warm lantern light"}
          ]
        }
      ]
    },
    "parameters": {"size": "1328*1328", "n": 1}
  }' | jq -r '.output.choices[0].message.content[0].image')
curl -s -o output.png "$url"

import os
import requests

response = requests.post(
    "https://api.nunchux.ai/v1/alibaba/services/aigc/multimodal-generation/generation",
    headers={
        "Content-Type": "application/json",
        "X-API-Key": os.environ["NUNCHUX_API_KEY"],
    },
    json={
        "model": "qwen-image-3.0-pro",
        "input": {
            "messages": [
                {
                    "role": "user",
                    "content": [
                        {"image": "https://example.com/street.png"},
                        {"text": "make it night and add warm lantern light"},
                    ],
                }
            ]
        },
        "parameters": {"size": "1328*1328", "n": 1},
    },
)

url = response.json()["output"]["choices"][0]["message"]["content"][0]["image"]
with open("output.png", "wb") as f:
    f.write(requests.get(url).content)

Tips

Set size explicitly. An omitted size renders at 2048*2048 and is billed at the 2K tier.
Keep a size at or under 2,250,000 pixels when 1K is enough. The common 1K sizes above all qualify.
Leave prompt_extend on for short prompts. Turn it off when the prompt is already detailed, or when you need the exact words rendered.
On an edit, put the images first and the prompt last. When you send several images, say which element comes from which image.
Download each image on receipt. The URL expires after 24 hours.

Errors and limits

n above 6 returns a 400 with the code invalid_request. So does a request with more than 3 input images. A refused request is not charged.
An error from Alibaba Cloud returns Alibaba Cloud’s own body, with code, message and request_id, for example InvalidParameter. The charge is refunded in full.
Qwen Image is priced per output image at the rate of its size tier, plus a small fee per input image. The rates are on the pricing page. Nunchux deducts the credits when it sends the request. When Alibaba Cloud returns fewer images than you asked for, Nunchux refunds the difference. See Credits and pricing.
A Qwen Image call counts toward your requests per minute. It does not take a simultaneous-job slot. See Rate limits.

The full error contract and the per-plan caps are on the Errors and Rate limits pages. How every partner model bills and refunds is on the Partner models overview.

Get started

Guides

Nunchux Optimized models

Partner models

Resources

Qwen Image

Qwen Image

Models

Endpoints

Request

Response

Example

Generate

Edit

Tips

Errors and limits

​Qwen Image

​Models

​Endpoints

​Request

​Response

​Example

​Generate

​Edit

​Tips

​Errors and limits

Qwen Image

Models

Endpoints

Request

Response

Example

Generate

Edit

Tips

Errors and limits