if I generate an image with a LoRA and then run it through Crystal Clear XL Lighting 27 times at 55% with something along the lines of "(corrected fingers:3)" to clean up the hands, fingers, etc. and then run it through the LoRA again, I often get a satisfactory finished result-- running it through the LoRA repeatedly is much less likely to clean up the fingers, feet, etc. (even with strong prompting)