How does training SDXL Checkpoint reduce deformed and poorly drawn hands?

One of the most frustrating problems I've encountered in training Checkpoint is how to make Checkpoint's hands look better instead of misshapen and knotted fingers or reduce the likelihood of misshapen fingers? I've tested on FLUX and the hand accuracy using Flux is much better than SDXL, but training Checkpoint on FLUX requires at least 40G of video memory, which is more than the average person can afford!