Image Generation

Image generation is a synchronous endpoint: a single request returns the result data[] directly. Both text-to-image and image-to-image (with reference images) are supported.

POST https://api.smartapi.cc/api/openapi/v1/generate/image/generations

Authentication#

Authorization: Bearer sk-your-key-here
Content-Type: application/json

Request Body#

Field	Type	Required	Description
`model`	string	Yes	Model id, e.g. `wan2.5-image`, `gpt-image-2`.
`prompt`	string	Yes	Text prompt.
`files`	array	No	Reference images (image-to-image); each has `file_uri` and `mime_type`.
provider config	object	No	Provider-specific config block, see table below.

Different providers accept different config blocks (match the model’s provider; send only one):

Provider	Config field	Options
QWEN	`qwenConfig`	`size`, `n`, `prompt_extend`
DOUBAO	`doubaoConfig`	`size`, `seed`, `guidance_scale`, `n`
GEMINI	`geminiConfig`	`aspect_ratio`, `resolution`
AZURE	`azureConfig`	`size`, `quality` (low/medium/high), `n`

To see which config block each model needs and the valid values (sizes, counts, quality, etc.) at a glance, open the “Image/Video Call Rules” tab on the Models page. You can also select the model in the Playground and toggle “Custom body / Debug” to inspect the exact request body.

Text-to-Image Example#

curl https://api.smartapi.cc/api/openapi/v1/generate/image/generations \
  -H "Authorization: Bearer $SMARTAPI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan2.5-image",
    "prompt": "A shiba inu running in snow, cinematic lighting",
    "qwenConfig": { "size": "1328*1328", "n": 1 }
  }'

Image-to-Image Example#

Pass reference image URLs in files to switch to image-to-image:

curl https://api.smartapi.cc/api/openapi/v1/generate/image/generations \
  -H "Authorization: Bearer $SMARTAPI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan2.5-image",
    "prompt": "Change the background to spring cherry blossoms",
    "files": [
      { "file_uri": "https://example.com/input.png", "mime_type": "image/png" }
    ],
    "qwenConfig": { "size": "1664*928", "n": 1 }
  }'

For the QWEN / Wan family (qwen-image, wan2.5/2.6/2.7-image, wan2.7-image-pro), image-to-image has a stricter pixel limit: total pixels must be between 768×768 (589,824) and 1280×1280 (1,638,400). 1328*1328, 1472*1140, 1140*1472 exceed the limit and will error in image-to-image — use 1664*928, 928*1664 or smaller. Text-to-image is not restricted.

Response#

{
  "data": [{ "url": "https://.../result.png" }],
  "model": "wan2.5-image"
}

Some models may return b64_json (base64 image) instead of url; handle both forms. If upstream validation fails, the endpoint returns the original upstream error message (e.g. size limit) to aid debugging.

Billing#

Image models are billed in credits (not tokens). FLAT pricing charges a fixed credit amount per image, so the total = base credits × count n. gpt-image-2 additionally applies a quality multiplier: low ×1, medium ×2.5, high ×5. See per-model prices in the Models page and the Billing doc for details.