Alibaba's Qwen Open-Sources Image Model Qwen-Image-2.1 with Native Transparent Image Support

On September 20, Alibaba's Qwen team announced the open-source release of Qwen-Image-2.1, the latest image model in its Qwen series. The model combines text-to-image generation and image editing in a single architecture, with only 7B parameters in its vision generation component, aiming to balance output quality with inference efficiency.

Key updates include native support for generating and editing transparent images, as well as layer-based editing and background removal. The model also supports up to 10 reference images to improve local editing and preserve fidelity for portraits and products. In addition, it brings enhancements to text rendering, lighting, and facial detail.

Qwen-Image-2.1 is now available as an open-source release via GitHub, HuggingFace, and ModelScope.