Skip to main content
POST

Supported Models

Prefer -pro for dense text/layout. Use qwen-image-3.0 for general images. Prompt max ~4.5k tokens. Model names are not interchangeable with the 2.0 series.

Authorizations

string
required
All endpoints require Bearer Token authenticationGet your API Key from the API Key Management Page:

Body

string
required
Model name
  • qwen-image-3.0 - Standard
  • qwen-image-3.0-pro - Pro (dense text / layout)
string
required
Text description for image generation, max ~4.5k tokens
string[]
Reference image URL array (image-to-image / editing), 1-3 images
  • Public http(s):// URL or data:image/png;base64,... (base64 is re-uploaded first)
  • Formats: JPG / JPEG / PNG / BMP / TIFF / WEBP / GIF
  • Per file ≤ 10MB; width/height recommended 384-2048 px
string
default:"1K"
Resolution tier (affects billing tier)
  • 1K (default; lowercase accepted)
  • 2K
You may skip the tier and pass pixels in size (e.g. 1600x900). With raw pixels, area > 2.25M pixels bills as 2K.
string
default:"1:1"
Aspect ratio, or explicit pixel sizeSupported ratios:
  • 1:1 (default)
  • 4:3 / 3:4
  • 16:9 / 9:16
  • 3:2 / 2:3
Also accepts 16x9 style, or pixels like 1024x1024. For custom pixels, each edge (width and height) must be 512–2048, aspect ratio 1:8 ~ 8:1.If neither size nor resolution is set, output is fixed at 1024×1024.
integer
default:"1"
Number of images, 1-6. Values above 6 are clamped to 6.
string
Negative prompt: content to avoid
boolean
default:"false"
Intelligent prompt rewritingDefaults to off for predictable results. Set to true to enable intelligent rewriting.
string
Rewrite mode (3.0 series only; requires prompt_extend: true)
  • direct - works for text-to-image and image-to-image
  • agent - text-to-image only, more aggressive rewrite
Invalid values are ignored.

Size Reference Table

Combine resolution (tier) + size (ratio):

Use Cases

Text-to-image (multiple)

Dense layout with rewrite

Image-to-image (edit)

Negative prompt

Limits

Response

integer
Status code; 200 on success
array
Response data array

Differences from 2.0

Notes

  1. Async: Submit returns a task_id. Poll Get Task Status every 3~5 seconds; client timeout ~3 minutes (2K + multi-image is slower).
  2. Storage: Generated images are mirrored to the platform CDN and remain available long-term.
  3. Billing: Per delivered image × resolution tier by actual pixel area (> 2.25M pixels = 2K). qwen-image-3.0 charges the same for 1K/2K; -pro 2K is 2× 1K. Failures are fully refunded; reference images are free.
  4. Errors: Custom size edge outside 512–2048, invalid ratio, agent on image-to-image, unreachable/oversized references → 400; rate limit → 429.
Query resultsUse Get Task Status for progress and result.images.