Alibaba's new open-weight model employs a unified checkpoint to handle both image generation and editing tasks. Qwen-Image-2.1 natively supports RGBA output for transparent images and allows the use of up to 10 reference images to guide visual output. The 7B parameter architecture is described by the company as outperforming most closed-source alternatives in quality and efficiency.

Integration for the system is available via SGLang-Diffusion for day-0 support of multi-image editing and text-to-image tasks. A live demo hosted on Hugging Face Spaces allows users to test the generation capabilities in a browser without local setup. This version improves visual quality for text and portraits compared to its predecessor while reducing inference time for multi-image inputs.