QwenCloud adds Qwen Image 2.1 Pro API access
The October 2 hosted API addition supports transparent images and editing with up to ten references.
Alibaba Cloud’s QwenCloud lists Qwen Image 2.1 Pro as an October 2 addition in its model release log. The hosted model supports image generation and editing, including transparent output and up to ten reference images.
The Qwen model repository dates the base Qwen Image 2.1 release to September 20, when downloadable weights and integrations became available.
The October entry concerns QwenCloud access. It does not establish a worldwide Pro debut.
Transparent output and multiple references
The release log says the prompt determines whether output has an alpha channel. That could let a designer place a generated subject over another background without a separate removal step, provided the model produces clean edges.
QwenCloud’s editing guide lists one to ten input images and one to six generated images for this model. Each input file can be up to 10 MB. Supported formats include JPEG, PNG and WEBP. Output is PNG, and only the first frame of an animated GIF is processed.
The integration details that matter
The QwenCloud API reference lists a recommended prompt maximum of 4,500 tokens and says excess tokens are truncated. Output area can range from 262,144 to 4,194,304 total pixels, with aspect ratios between one to eight and eight to one. If size is omitted, the model chooses a resolution from the prompt.
Prompt rewriting is enabled by default. Direct enhancement is the default mode, while agent enhancement is available for text to image generation. Thinking is also enabled by default and requires prompt rewriting. The documentation says it can improve image quality at the cost of additional generation time. These are documented controls and vendor claims, rather than independently measured performance.
Alibaba Cloud’s own API reference also confirms the Pro model’s support for ten references and transparency. It documents an important integration detail. Although the request options follow the 3.0 family, the usage response reports image_count, width and height. A parser built around the 3.0 family’s output_image_count and output_width fields needs to account for that difference.
Generated image links expire after 24 hours, according to the editing guide. Applications should save finished files rather than treat those temporary links as permanent storage. For an existing integration, reference handling, response parsing and file retention are practical checks before rollout. Output quality and latency still need testing on the images the team actually uses.
ByteForward’s earlier coverage of Qwen 3.8 Max API access similarly separates a hosted service release from plans for downloadable weights.
Illustrative editing workspace photographed by Jakub Zerdzicki on Pexels under the Pexels License. The photograph does not show QwenCloud.
