This LoRA is used to create high quality alpha masks from either an outlined entity (drawn or rectangular), or a description of the entity. Unlike SAM, this solution will emit grayscale masks, good for use in compositing. Unlike Qwen Image Layered, this process can be used to create masks of arbitrary resolution, using the qwen image edit 2511 model

Trained a new model with a much larger dataset. Specifically focusing on single images with different instructions to mask different elements from the image. Much better instruction following.

Example prompts:

This LoRA was trained using this dataset, which was created using this toolset.

You should use Create a black and white alpha mask of the ... to trigger the image generation.

Download them in the Files & versions tab.