Qwen-Image-Edit: Image Editing with Higher Quality and Efficiency

Qwen-Image-Edit combines two representations of an input image: a vision-language model supplies information about its meaning, while an image encoder supplies information about its appearance. According to the official introduction, this supports both changes that preserve a subject's identity and localized edits to visual details.

Those goals require different kinds of preservation. A style change may alter most pixels while keeping a character recognizable; changing lettering instead calls for retaining surrounding appearance. The team also describes English and Chinese text editing that aims to preserve the original lettering's font, size and style. This makes the intended scope of an edit central to judging whether the result succeeded.

“Qwen-Image-Edit simultaneously feeds the input image into Qwen2.5-VL (for visual semantic control) and the VAE Encoder (for visual appearance control), achieving capabilities in both semantic and appearance editing.”