Qwen-Image-2.1-Pro is a unified text-to-image generation and image-editing model in the Tongyi Qianwen (Qwen) series. Its visual generation component contains only 7 billion parameters (a 32-layer single-stream DiT), achieving a good balance between generation quality, inference efficiency, and versatility. This release includes four key improvements: - Compact and efficient: adopts a lightweight architecture that combines hybrid-granularity attention and prefix KV cache reuse to deliver high-quality image generation with low computational cost. - Native transparency support, unified generation and editing: can generate regular or transparent (RGBA) images from text, edit transparent layers, and extract subjects from photos—all handled by a single model. - Versatile editing: supports up to 10 reference images, allows specifying local edit regions via circular or freehand annotations or separate masks, and preserves identity features of people and products. - Photorealistic textures and refined aesthetics: improved typography, portrait lighting, and detail rendering for more visually appealing results.