Support my work by buying me a coffee! Tossing even just a couple dollars my way will help me fund more Qwen Image Edit LoRAs. Thank uuuu ~~ <33

This LoRA (designed for the 4-step Qwen Image Edit Lightning LoRA for fast inference) creates a black and white matte (layer mask) of the prominent subject of the image. Once generated, go into your favorite image editing software (eg. Photoshop, GIMP, etc) and apply the output matte as a layer mask to fully remove the background of the subject. This is robust matting, meaning that it works across a wide variety of scenarios such as matting multiple subjects in the image, matting entire comic panels, mattes complex hair, among additional scenarios. Works on art, real pictures, animals, etc.

AI matting will never be perfect however, so on occasion you will have to manually adjust the mask by painting in with white or black on the layer mask, or by using curves on the mask. There may also be color fringing as well after the mask is applied, as that is the nature of matting. You'd have to manually decontaminate the colors yourself.

In order to use this, use it at strength: 1, CFG 1 @ 4 steps with Qwen Image Edit Lightning LoRA, and just input the following :

Select the subject of the image. The subject should be fully white, while the background should be fully black.

This was only trained on selecting the prominent subject, so it can't select other parts of the image that you specify.