Qwen's Image 2.1 PE I2I turns an editing request and reference images into instructions for an image editor. The model examines the supplied images alongside the request, then returns a rewritten prompt and information about the intended image proportions. Its model card shows that text being passed to the separate Qwen Image 2.1 model, which performs the edit.
This separates interpreting the requested change from rendering it. A developer can inspect or modify the rewritten instruction before generating an image, including whether the output should retain a reference image's proportions.