Qwen-Image-2.1 is Alibaba’s open-source image model in the Qwen family, balancing generation quality, inference efficiency, and cost. It unifies text-to-image generation and image editing in a single model, with just 7B parameters in its visual generation component and native support for generating and editing transparent images.
The model uses a lightweight architecture of 32 Single-Stream DiT layers, with a mixed-granularity attention design and KV cache reuse for faster inference and lower memory use, even when editing with multiple input images. Editing is versatile: it supports up to 10 reference images for multi-subject compositions, virtual try-on, and interior design; local editing via circles, painted annotations, or separate masks; and strong fidelity that preserves portrait identity and product consistency across edits. Task coverage spans panoramas, infographics, and storyboards.
Qwen-Image-2.1 further improves visual quality with realistic textures and refined aesthetics, including better typography, portrait lighting, and fine details. Its native transparency lets it generate regular or transparent images from a prompt, edit transparent layers, and extract subjects from photographs as RGBA layers for reuse in design work. Released open-source with weights on GitHub, Hugging Face, and ModelScope, it is a practical tool for design, content creation, e-commerce, and visual storytelling.
Get more likes & reach the top of search results by adding this button on your site!
Get top updates in AI to your inbox every weekend. It's free!