@johnsmit
Qwen just made AI images editable by default with Qwen-Image-Layered
Qwen-Image-Layered is a new diffusion model that takes a flat RGB image and decomposes it into multiple semantically separate RGBA layers — background, characters and objects all live on their own layers. Each layer can be moved, resized, recolored or removed without touching the rest of the image, which largely eliminates the usual artifacts from re‑generating everything.
The model supports variable layer counts and even recursive decomposition, so you can start with a simple split (subject vs background) and then further break down complex layers into clothing, face, hands and more. You can also steer it via text by asking for 3, 5 or 8 layers and letting the model adapt the structure to your scene.
Instead of giving you a single pretty but “baked” image, Qwen-Image-Layered outputs something closer to an auto‑built Photoshop project.