Qwen Image parameters and API differences
Reference images, output sizes and prompt enhancement for Qwen Image.
See Create / edit images for the OpenAI-compatible request structure.
Applicable models
| Model | Supported APIs |
|---|---|
| qwen-image-3.0 qwen-image-3.0-pro | Generation and editing through the OpenAI-compatible API |
| qwen-image-2.0 qwen-image-2.0-pro | Platform-adapted generation and editing through the OpenAI-compatible API |
3.0／3.0 Pro
| Field | Differences |
|---|---|
image | Editing supports 1–3 reference images, referenced in array order. Omit for text-to-image generation. |
size | Use WIDTHxHEIGHT or auto. If omitted, the model chooses. Pixel area is 512×512–2048×2048, with an aspect ratio of 1:8–8:1. |
prompt_extend, prompt_extend_mode | Enhancement is enabled by default. Default mode direct supports generation and editing. agent is for text-to-image generation only. |
enable_thinking | Enabled by default. Applies only with prompt_extend: true. |
negative_prompt | Both models let you specify unwanted image content. |
2.0／2.0 Pro
The following table is based on Alibaba Cloud's generation and editing documentation for the 2.0 series.
| Item | Differences |
|---|---|
| Reference images | Editing supports 1–3 references as URLs or Base64 Data URLs. |
| Generation size | Defaults to 2048×2048, with a pixel area from 512×512 to 2048×2048. |
| Editing size | If omitted, pixel area is approximately 1024×1024 and the aspect ratio follows the input image. With multiple inputs, the last image determines the ratio. |
| Output | Both models output PNG and generate 1–6 images per request. |