Reference-guided inpainting with Qwen Image Edit. Learn more about the Qwen-Image series.

How to use: draw a mask on the base image to mark the region to replace, upload a reference image holding the thing you want placed there, then describe it in the prompt (e.g. "a coffee mug on the table, blended naturally with matching lighting"). The reference is scaled to fit inside your masked area and pasted in, then the model blends it. Use Strength ~0.6–0.85 so the reference survives the inpaint pass — near 1.0 the model mostly ignores it. The reference image is optional; leave it empty for plain text-driven inpainting (then 1.0 is fine).

This demo uses the Qwen-Image-Lightning LoRA for accelerated 8-step inference. Try on Qwen Chat, or download model to run locally with ComfyUI or diffusers.