Made an attempt at cleave gags, they're alright. Separated into two Loras, one for normal and another for knotted. The latter had a much smaller dataset and generally performs worse especially when it comes to texturing the knot, but there's at least a noticeable difference between the two. I tagged them at the same time then trained separately, so they share a trigger word.
Trigger Word:
clvI removed all "gag"-related tags since I figured they'd interfere, but I have found that adding
cleave gagactually improves accuracy in some cases, at least when using a bondage model. Give that a try as well if you find it's not being consistent!
I also had some instances where it was noticeably difficult to add bindings with some combinations of other Loras and I'm unsure why. Perhaps this is better suited to inpainting? I also found the knotted cleave model to be more overfit than the normal one, so I had to use a smaller CFG to compensate. It also sometimes added animal ears, no clue why.
Recommended weight: ~1.0
Description
Trained on animefull-final-pruned on ~1100 images, some of which are cropped from larger ones. 1 repeat, trained for 10 epochs on batches of 3, used epoch 9.
FAQ
Comments (5)
Dude... how can it be so good? ❤️
Where did you get the base model for your sample pics? storeBoughtBondage_v1?
It's a mix of
Store Bought Gyoza Mix + bondage-v11 + v1-5-pruned-emaonly
Multiplier: 0.5
Add difference, copy config from A, B, or C
The first two are on Civitai, third is here https://huggingface.co/runwayml/stable-diffusion-v1-5
Hi. Great model, but how do I use it? It is not shown when I searched it on additional resources.
It's just a regular Lora, so you just need to add the trigger word and the call to your prompt. But if you're talking about the model itself, those are just personal mixes I made, which I mention in the description here: https://civitai.com/models/55923/tape-gag For what it's worth though, I stopped doing that in favor of using this (https://civitai.com/models/89556/bondage-v11-lycoris) at a low weight (~0.2) with any given model. I've found it affects the resulting style a lot less.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.