curl --request POST \
--url https://api.tikway.ai/v1/images/generations \
--header 'Authorization: Bearer <token>' \
--header 'Content-Type: application/json' \
--data '
{
"model": "bailian/qwen-image-3.0-pro",
"prompt": "A cinematic street scene in Shanghai on a rainy night, with neon lights reflected on the wet road, realistic photography",
"size": "1024x1024",
"prompt_extend": true,
"watermark": false
}
'{
"created": 1788451200,
"background": "opaque",
"data": [
{
"b64_json": "iVBORw0KGgoAAAANSUhEUgAA...",
"index": 0
}
],
"output_format": "png",
"size": "1024x1024"
}Image generation
qwen-image-3.0-pro (as well as the bailian/ alias) supports wensheng images and tsunimages of up to 3 reference images on the same endpoint.
Supports: model, prompt, size, images, prompt_extend, prompt_extend_mode, enable_thinking, negative_prompt, seed, watermark, response_format.
The official compatible interface also supports n=1..6, but the current gateway will explicitly reject Qwen’s n, so this specification does not expose this field. prompt_extend_mode=agent only supports Vincentian diagrams; enable_thinking only takes effect when prompt_extend=true.
curl --request POST \
--url https://api.tikway.ai/v1/images/generations \
--header 'Authorization: Bearer <token>' \
--header 'Content-Type: application/json' \
--data '
{
"model": "bailian/qwen-image-3.0-pro",
"prompt": "A cinematic street scene in Shanghai on a rainy night, with neon lights reflected on the wet road, realistic photography",
"size": "1024x1024",
"prompt_extend": true,
"watermark": false
}
'{
"created": 1788451200,
"background": "opaque",
"data": [
{
"b64_json": "iVBORw0KGgoAAAANSUhEUgAA...",
"index": 0
}
],
"output_format": "png",
"size": "1024x1024"
}Authorizations
Bearer authentication header of the form Bearer <token>, where <token> is your auth token.
Body
Required. It is recommended to use the public model ID bailian/qwen-image-3.0-pro; the short model name without the vendor prefix is only available if the account alias is configured.
bailian/qwen-image-3.0-pro, qwen-image-3.0-pro Required. Image generation or editing instructions support Chinese and English. The official recommendation is no more than about 4,500 tokens.
1Optional. Output size in the format WIDTHxHEIGHT, for example 1024x1024. When auto is omitted or passed, the gateway sends 1024x1024 to the upstream; the official pixel area range is 512×512 to 2048×2048, and the aspect ratio is 1:8~8:1.
^(auto|[1-9]\d*x[1-9]\d*)$"1024x1024"
"2048x2048"
"1536x1024"
Optional. Drawing/editing reference pictures, up to 3 pictures. Each item provides a file_id or image_url; when omitted, a Vincentian image.
1 - 3 elementsOptional. Whether to enable intelligent prompt word rewriting.
Optional. Prompt word rewriting mode. Direct supports Vincentian diagrams and Tusheng diagrams; agent only supports Vincentian diagrams.
direct, agent Optional. Whether to enable thinking mode. When set to true prompt_extend cannot be turned off explicitly.
Optional. Describe what you don't want to appear in the results.
Optional. Random seed. Fixed seeds improve result reproducibility but do not guarantee pixel-level consistency.
0 <= x <= 2147483647Optional. Whether to add Qwen-Image watermark in the lower right corner of the generated image.
Fixed returning Base64 image so that the response conforms to the Response Schema of this interface.
b64_json, url Response
The image was generated successfully. The gateway has downloaded the upstream temporary image and converted it to Base64 without the data URL prefix.
The Unix timestamp (in seconds) when the gateway completed response conversion.
x >= 0The current Qwen path is fixed to return opaque.
"opaque"The format inferred by the gateway based on the Content-Type of the downloaded image. Omitted when no recognized type is provided by the upstream.
png, jpeg, webp Actual or requested output size. Omitted when neither the request nor the upstream response provides a size.
^[1-9][0-9]*x[1-9][0-9]*$
