Intended to depict bondage mittens often used in pet play and other bondage images.
VERSION 2
---------------------------------------------------------------------------
Strength: 1.2-0.8
Trigger Prompt:
, Bondage, (BGBondageMitts, BGPetPlay, :1.3), (<leather/latex> bondage mitts, :0.8),
Trained Terms:
BGBondageMitts -- trigger word this is required
BGPetPlay -- "pet play poses" (mostly collar and leash) it doesn't really work without this term so it's basically a required trigger word.
BGPawGloves -- the bondage mittens will be paw-shaped and will have paw-pads (you can use this instead of BGPetPlay actually in your prompt)
BGAllFours -- pet play pose - crawling on all fours
BGBegging -- Pet play pose 'begging' -- hands up in a begging pose (often kneeling but not always)
BGPiddleFours -- Crawling on all fours -- but in a Bitch suit...rule of thumb if you want crawling use this if you use BGBitchSuit otherwise use BGAllFours
BGBitchSuit -- more of a half bitch suit, the legs will be in a leg binder but the hands will be "free" in bondage mittens -- it's the best thing this lora can generate in my opinion.
BGMaid -- There's lots of maid themed bondage mittens -- this produces a maid dress
BGBalletHeelBoots -- ballet pointe style boots for your maid (or pet in general)
VERSION 1
---------------------------------------------------------------------------------------
Use at: 0.8-1 strength
Trigger Words: bondage gloves, (bondage mittens, bondage mitts, :1.2),
Optional Extras:
latex bondage mittens - for shiny mittens (it tends to do this even if you don't ask for it)
Paw Mittens - for paw pads (usually pink) on the "palms" of the gloves.
Bondage Socks - doesn't always work but may create similar socks to match the mittens.
Usage Note: if you don't put enough weight on the model or on the prompt it may sometimes try to draw in the thumbs which is the very thing I designed this model to NOT do, but that's what the base Illustrious model thinks a bondage glove should look like.
one of my primary complaints about "some" lora creators (not all) is simply that they don't use enough images in their training which tends to result in the model learning certain traits of their images or all the images tend to be of a single character or in a single style or from a single angle -- I try to avoid that in the lora I create and I think the results speak for themselves. I also create my own data sets from hand picked images that are tagged by hand (which is why I haven't released a ton of models-it's very time consuming).
Description
This version was a pain to create and I'm not TOTALLY happy with it but it does work pretty well most of the time.
------------------------
The making of this Lora -- The Saga
So I've been working on this for almost 3 months now...this is version 2.1 cause I didn't like 2.0 at all I threw the whole thing away and started over. Illustrious has a VERY VERY strong desire to draw fingers and thumbs...and NORMALLY this is good...it results in less deformities in the hands in general...but to overcome that this lora needed a lot more training than your traditional lora -- which meant I had to change my learning settings and train more epochs and to do that without overlearning meant I also needed a lot more images in the data set.
The result -- well I do see some style changes that are typical of overtraining but honestly I tried this in several styles and it works fine it does tend to produce more detail because many of the images are higher resolution than the default Illustrious resolution.
I WANTED originally to have the mittens work with just 1 word no extra emphasis added -- BGBondageMitts -- and about 1/3 of the data set uses just that 1 word and nothing else so I expected that to work but about half the data set also is pet-play themed...so that ended up being too strong -- short version -- it typically only works if you use the BGPetPlay prompt (as indicated in how to use). But in short the data set is more than double the size of what it was for v1 (which was already over 100 images) and I think it produces generally better images.