If you like this and wanna throw over some support, buy me a coffee.
This took way too long. I've had the pose down pretty quick but the feet details gave me so much trouble.
As you can see from the version number, it's been a looong ride. I decided to release it, it's not quite as good as I'd like but it has come a long way. Will occasionally do body horror still, add too many feet or toes, so it might take a few attempts.
Best way is to start with a rough description like
woman lying down <lora:restfeetup55_rvkwi:1>, soles, YOURDETAILS
There's no single keyword to trigger it, as i had to throw off the text encoder badly all over the place in training (because it kept ruining feet) to give the UNET more time to grasp them better visually.
It should be flexible in details you specify and listen to the prompt well overall.
I also trained it in a bunch of front camera angles and rotations, try wide vs tall resolutions to ???????? different looks (more of a side view). Avoided these here as the tall images make for better previews.
It's trained on SD1.5, tested in URPM 1.3 and grapefruit v4. Should work in most models that have somewhat of an understanding of anatomy.
PS: I made a negative embedding for grapefruit while switching between models/testing, it's not needed for this pose to work at all. But it is needed to recreate those preview images 1:1, you can grab it here.
