It uses MobileSAM for segmentation, which works great for text on CG images.
Download the models and unzip the SAM folder under ImageTrans's root: https://github.com/xulihang/MobileSAM/releases/download/models/SAM.zip.
Then you can use the SegmentAnything Mask Generation plugin to use it.
You can start a Python server according to this repo: https://github.com/xulihang/MobileSAM.