Alibaba's AI division, Qwen, has unveiled a new image editing model that breaks photos down into editable, independent components. Named Qwen-Image-Layered, the model separates images into multiple layers with transparent backgrounds (RGBA layers), enabling users to modify individual elements without affecting other parts of the image.

The model supports basic editing tasks such as resizing, repositioning, and recoloring specific elements. Users can swap backgrounds, replace figures, edit text, or remove, move, and enlarge objects within the image. The system allows decomposition into either 3 or 8 distinct layers, and the process is iterative—each layer can be further subdivided as needed. The Qwen team describes this approach as bridging the gap between standard image formats and structured, editable representations.

The Qwen team has released the code on GitHub, and the model is available on both Hugging Face and ModelScope. Additional details can be found in the accompanying blog post and technical report. Live demonstrations are also accessible on Hugging Face and ModelScope for hands-on testing.