Why not just enlarge the images while preprocessing in textual inversion?

[YYZ252]

When I used the "preprocess" before "training", I found the "split oversized images" checkbox confusing. If the images is not in 1:1 ratio, it will clip the images into parts, the products images often loses some parts. If the person at the bottom of the image, it may be cutted out. I think it may make the training result worse. I wonder if it works to just enlarge the image rather than clip it. For example, if the image's resolution is 1000512, we can first enlarge the image to 10001000, then resize it into 512*512 (the default resolution, but I wonder if the training result will be better if I set a larger resolution). Then we also get a square image but never lose the important parts.