Update V2: Expanded Dataset: now with or without ball gags. Updated trigger.
v2a: better tension
v2b: better variety/consistency
This is a fairly difficult Pose even for Pony XL to understand. I highly recommend an image to image workflow as described here here.
Sample Outputs above are straight text to image though, so you can get results with this as well.
If you love my work: Consider supporting the cause ;)
This is a retrain of this SDXL Model
It is heavily biased towards realism. If you don't want that you have to heavily prompt against it and/or use style lora.
If you can't get good results with txt2img with this model: Try the following steps:
-change your checkpoint. Some have an easier time with this pose than others
-clean up your prompt. try getting the pose to work first and then selectively add your desired elements back in.
-clean out your other lora: some of them may fight with this model and none of them will win.
-as always: try higher steps, lower cfg. Upscaling will probably not save you on this one though. Don't Upscale before you have a decent base.
-use some of the tags below
Taglist
To get the model to understand what's what, some pretty extensive manual tagging was necessary. Now the lora has a pretty long list of charged tags, that you may use (with varying effectiveness) in positive or negative: (don't go overboard though;)
Perspective: From above | from side | from angle | facing viewer -- these may also be combined
(From behind confused the model to much so it was removed from the dataset)
Setting: On floor | on ground | on table | on bed ...
Pose: arched back | head down | head back | feet crossed | feet together | knees together | knees apart feet together | elbows together | hands together elbows apart
Bondage related: (color) + gag | (color) + rope | natural rope | feet linked to hands | feet linked to shoulders | feet linked to hair | bound + (bodypart)