Alibaba has released the Qwen-Image-2.1 model, which combines text-to-image generation and image editing capabilities. The model features a lightweight architecture with 7 billion parameters, designed to balance generation quality with computational efficiency.
A key feature of the new model is its native support for generating and editing transparent images. Users can create transparent layers directly from text prompts or perform background removal tasks using the integrated editing tools.
The model supports the use of up to 10 reference images to improve local editing, portrait fidelity, and product rendering. This functionality allows for complex tasks such as interior design visualization based on multiple input samples.
Alibaba has also improved the model's performance in text rendering, lighting effects, and character detail. The project is now available for developers on GitHub, HuggingFace, and ModelScope.
