>>41130885This one didn't come out quite as Asanagi looking.
The AI does not know really feet. High denoising just leads to monstrosities. Low denoising is a matter of generating the foot 20 times and picking the least deformed one. Then doing it again, and again, and again. And the closer you get to the foot looking human the less likely it is for any of the next 20 iterations to be an improvement.
Base image for comparison:
https://litter.catbox.moe/llfwf9.png 