This is a Lora for the "bound to bed" concept, with legs and arms spread and tied to the corners of the bed with ropes.
To use it, I propose a structure like the following:
[photograph/comic/etc]... of a woman, lying on ... with her arms and legs spread and bound with [color] rope. Her wrists are bound to the ... with ropes. Her ankles are bound with ropes and tied to the bed's corners. The ropes connecting her ankles and wrists to the bed frame are tight. She is gagged with a piece of duct tape in her mouth. She is wearing ...
But... this is just a starting prompt and like all FLUX LoRAs, if you explore it, you can create more poses (see samples).
Plus:
The LoRA can create a tape gag (of any color)
If you want another type of gag, better use another LoRA. I tried including ball gags and cleave gags, but I guess the data set (98 images) was not enough (or me)
Use any color of rope.
The LoRA has knowledge of crotch rope (!) and sometimes of breasts bound in shibari (not that easy)
See the samples for more...
(I know there's another pretty good bound to bed LoRA, but I wanted emphasis on ropes)
Description
FAQ
Comments (5)
out of curiosity: whats your dataset size? and did you limit it to one perspective? i have one that works like shit...
Hi! 94 (Had to check it). There are different view points, but all are depicting a woman spread on the bed (no, I think there are some with her legs joined). But, I haven't prompted for a viewpoint (might do that in another run, if I expand the set a bit)
@zkrd maybe ive just overtrained. i lately had better luck training to 4000-8000 steps with poses but yours works better than mine with 1397. Thanks for the answer anyways ;)
@Joschek IMHO it depends on the pose. For the hogtied and heels LoRA I went for 3750 steps in 30 epochs with a set of close to 400 or so. I also tag using joycaption, then edit each description one by one to keep consistency. I found that using a standard phrase for the description helps.
@zkrd yeah my hogtied lora is similar in size and under 200 it worked like shit, i use standard sentence start + 1 or 2 cogvlm sentences + danbooru tags added if its difficult for flux. but sometimes standard phrase + danbooru is enough (i do believe some of the danbooru tags are helpful for detail prompting but cant really prove it)