Skip to Content
ModelsQwen-Image

Qwen-Image family

Alibaba’s Tongyi image models, two in total, called through the OpenAI-compatible endpoint /v1/images/generations; both text to image and editing use this endpoint.

Choosing a model

The two models take exactly the same parameters. Pro is the premium version, with better text rendering and image quality.

Model IDs (copy them exactly)
Qwen-Image 3.0 Proqwen/qwen-image-3.0-proPremium tier
Qwen-Image 3.0qwen/qwen-image-3.0Standard tier

For live prices, see the model page .

Differences from other families

Key pointReturns an image link rather than base64, and the link expires after 24 hours.
Qwen-ImageGPT Image, Seedream, etc.
Returnsdata[].url, a temporary Alibaba Cloud OSS linkdata[].b64_json
Reference image fieldinput_imagesSeedream uses image
/v1/images/editsNot supportedSupported by GPT Image

Link lifetime: the Expires parameter in the link is 24 hours after generation. For long-term use, download and store the image promptly; do not save the link directly to your database.

Text to image

qwen_image.py
import urllib.request
from openai import OpenAI
 
client = OpenAI(api_key="YOUR_OFOX_API_KEY", base_url="https://api.ofox.ai/v1", timeout=600)
 
resp = client.images.generate(
    model="qwen/qwen-image-3.0-pro",
    prompt="A red apple on a white table, next to a card that reads “OfoxAI”",
    size="1024x1024",
)
 
# The response is a temporary link; download it within 24 hours
urllib.request.urlretrieve(resp.data[0].url, "output.png")

Size and image count

Key pointTotal pixels must be between 512×512 and 2560×2560; n is at most 6.
ParameterValueNotes
sizeWIDTHxHEIGHTTotal pixels from 262,144 (512×512) to 6,553,600 (2560×2560); out of range returns 400
size omitted—Measured output 2048×2048
n1–6Returns several links in one call; 7 or more returns 400

2560x1440 works in our tests. For latency, 1024×1024 takes about 10 to 60 seconds, 2560×1440 about 60 seconds, and editing about 50 to 100 seconds. Set the client timeout to 300 seconds or more.

Image editing & multiple reference images

Key pointReference images go in input_images, up to 3; the image field is not used, and /v1/images/edits is not supported.
qwen_edit.py
import base64
import urllib.request
from openai import OpenAI
 
client = OpenAI(api_key="YOUR_OFOX_API_KEY", base_url="https://api.ofox.ai/v1", timeout=600)
 
with open("apple.jpg", "rb") as f:
    ref = "data:image/jpeg;base64," + base64.b64encode(f.read()).decode()
 
resp = client.images.generate(
    model="qwen/qwen-image-3.0-pro",
    prompt="Make the apple purple, keep everything else the same",
    extra_body={"input_images": [ref]},
)
 
urllib.request.urlretrieve(resp.data[0].url, "apple_purple.png")
  • input_images is an array; each item is an image URL or a string starting with data:image/...;base64,, with 1 to 3 images. More than 3 returns 400.
  • image is a Seedream-only field; Qwen-Image does not support it.
  • When editing without size, we measured an output of 2048×2048, which may not match the original. Pass it explicitly to keep the size.

Other parameters

negative_prompt (content you don’t want to appear) and prompt_extend (whether the model automatically expands the prompt) can be passed and work in our tests. For parameter definitions, see the Alibaba Cloud Model Studio official docs .

Common errors

Error message (excerpt)CauseFix
Image area must be between 262144 (512x512) and 6553600 (2560x2560) pixels for t2i requestsSize too small or too largeAdjust to within range, e.g. 1024x1024 or 2560x1440
Input should be less than or equal to 6: parameters.nn is above 6Use 6 or fewer; for more images, send requests in batches
supports 0~3 image content items. Got 4 image items.More than 3 reference imagesReduce to 3 or fewer
does not support the /v1/images/edits endpointThe edit endpoint was calledUse input_images on /v1/images/generations
Reference image has no effectThe image field was usedUse input_images
Image link cannot be openedThe link is past its 24-hour lifetimeGenerate again; download and store images promptly after getting the link

qwen-image-3.0-distill in the error messages is the upstream internal model name; keep calling with the model IDs listed on this page.

Exact error messages

Image area must be between 262144 (512x512) and 6553600 (2560x2560) pixels for t2i requests. Got area=10000 (width=100, height=100).
Input should be less than or equal to 6: parameters.n
Model 'qwen-image-3.0-distill' supports 0~3 image content items. Got 4 image items. (0 images = T2I mode, 1~3 images = I2I mode)

FAQ

Why does Qwen-Image return url instead of b64_json?

Qwen-Image returns a temporary Alibaba Cloud OSS image link (data[].url), not base64. The link expires after 24 hours, so download and store the image as soon as you get it.

How do I edit images with Qwen-Image?

Pass input_images on /v1/images/generations (an array of 1 to 3 items, each an image URL or base64 starting with data:image). image is a Seedream field that Qwen-Image does not support, and /v1/images/edits is not supported either.

Qwen-Image returns Image area must be between 262144 and 6553600 pixels. What should I do?

The total pixel count must be between 512x512 and 2560x2560. For example, 100x100 is too small and 3000x3000 is too large. Use 1024x1024, 2048x2048, 2560x1440 or similar.

How many images can Qwen-Image generate at once?

n is at most 6; 7 or more returns 400. For more, send requests in batches.

Why can I no longer open a Qwen-Image link?

The Expires parameter in the link is 24 hours after generation; after that the image cannot be downloaded and must be generated again. Download the image promptly after getting the link and save it to your own storage.

Last updated on
Seedream (Doubao)MAI Image