Skip to main content

​
Bailian Qwen-Image series

Generate images with Tongyi Qwen Bailian Qwen-Image models. Supports prompt refinement and strong Chinese-style semantics.

​
Basics

  • Endpoint: POST /v1/images/generations
  • Auth: Bearer Token
  • Content-Type: application/json

​
Model overview

ModelSummaryOutput
wan2.6-t2i **recommended**Wanxiang 2.6—free size within total pixel and aspect constraints (same idea as 2.5)Resolution: total pixels in [1280×1280, 1440×1440]; aspect [1:4, 4:1]; PNG
wan2.5-t2i-preview **recommended**Wanxiang 2.5 preview—flexible sizing within pixel/aspect bounds (e.g. 768×2700; 2.2 and below cap single side at 1400)
wan2.2-t2i-flashWanxiang 2.2 flash—~50% faster than 2.1Resolution: both sides in [512, 1440]; PNG
wan2.2-t2i-plusWanxiang 2.2 plus—better stability vs 2.1
wanx2.1-t2i-turboWanxiang 2.1 flash
wanx2.1-t2i-plusWanxiang 2.1 plus
wanx2.0-t2i-turboWanxiang 2.0 flash

​
Request body

FieldTypeRequiredDescription
modelstringYese.g. qwen-image-plus or qwen-image-max.
contentstringYesImage description text.
parametersobjectNoGeneration controls.
└─ negative_promptstringNoThings to avoid.
└─ prompt_extendbooleanNoAI prompt expansion (default true).
└─ watermarkbooleanNoAdd watermark or not.
└─ sizestringNoe.g. 1328*1328, 1024*1024; width and height each 512–1440.
└─ nintegerNoNumber of images, 1–4, default 4. Billed per image—use 1 for tests.
└─ seedintegerNoSeed in [0, 2147483647]. Same seed tends to stabilize output but is not a hard guarantee.

​
Response

200 OK
FieldTypeDescription
createdintegerUnix timestamp.
dataarrayResults.
└─ urlstringDownload URL.
└─ b64_jsonstringBase64 if requested.
└─ revised_promptstringPrompt after AI refinement.

​
Request example

{
    "model": "wan2.2-t2i-flash",
    "prompt": "A kitten running on grass, cinematic look, natural light and shadow",
    "parameters":{
        "size": "960*1390"
    }
}

​
Response example

{
  "created": 1713833628,
  "data": [
    {
      "url": "https://example.com/qwen_output_image.png",
     	"b64_json": "",
      "revised_prompt": ""
    }
  ]
}

​
Tips

  1. Chinese semantics: Qwen is strong on idioms, couplets, and Chinese motifs (e.g. Yueyang Tower, blue-and-white porcelain).
  2. Prompt extend: Keeping prompt_extend: true usually yields richer detail.