Works with natural language prompts. Primarily trained on hogtie poses, but it has learned bondage more conceptually and can generalize into other restraint setups like chair ties and similar poses.
All example videos are generated with txt2va only, with turbo, around strength 0.8. Has worked for me well on all three - txt2va, flf2va, and ref2va
Strength: 0.8-1.0 works great. Can be pushed higher if needed; this LoRA stays fairly stable even above 1.0.
Trigger word: not required.
Main hogtie prompt structure:
"Woman is lying on a [surface] bound in a hogtied pose with her arms pulled behind her back and her legs bent upwards."Rope / restraint positioning:
"She is bound with [color] [material:rope, belts etc.], with her [wrists/elbows/arms/ankles/thighs/knees/chest/shoulders] tied/restrained [position]. Her ankles are tied to her bound [wrists/hair/etc.]."Describe the exact restraint placement you want in natural language.
It can handle different rope colors/materials and different body areas.
It can also generalize into non-hogtie restraint setups, especially chair ties and similar bondage concepts.
Gag Options:
Don't include anything if you don't want a gag.
Ball gag:
"She has a [color] ball gag in her mouth fastened with [color] [material: leather, etc.] straps."OTM Cloth Gag / Wraparound Tape Gag:
"She is gagged over-the-mouth with [color] [tape/cloth] wrapped around her mouth fastened behind her head." // or "She has a [color] [material: tape/cloth] wrapped tightly around her mouth [looping/fastened] behind her head."Cleave Gag:
"She has a strip of [color] cloth cleave gag between her lips fastened behind her head.
Motion / Audio prompting:
Primary trigger for struggling noises/sounds can be described as "muffled vocalizations" or "muffled sounds".
Describe movement naturally after the restraint setup. For example:"She writhes, arches her back, tilts her head, and struggles against the restraints."For speech-like gagged audio, try:
"She attempts to speak through the gag, producing muffled speech-like vocalizations with conversational rhythm and cadence."Other subjects:
Works with men and multiple subjects too. For best consistency, describe each subject separately rather than interweaving their restraint descriptions.Describe the subject in natural language.
Describe the scene/environment in natural language.
Describe camera angle and lighting normally.
Note: Hogtie is still the strongest / most consistently learned concept. The broader bondage behavior is a useful side effect of the training, so I would not claim this as a full general-purpose bondage LoRA, but it can do a lot of fun things as you can see in my examples.
Description
FAQ
Comments (21)
Special thanks to @nofmegan895 for all his testing help to get me to the final release!
Thank you, also for this great LoRA this final release work very well and worth the testing with you
This works really well!
Thank you for the feedback!
Finally... Do you think we'll see any other bondage LoRAs coming along? Like a dedicated mummification LoRA?
I'd be shocked if there weren't more BDSM related loras coming. Minimax H3 is able to do a lot of BDSM related stuff by default. So it only makes sense for more loras to come out for it than other models.
@RenegadeZebra It's already been a month and I'm pretty sure this is the first one. I've had trouble getting it to do BDSM things, everyone has been so focused on the more hardcore or full on porn that it feels like the softcore stuff is getting forgotten or ignored.
Well, download what you can, e.g. the Whip - Minimax H3 - Lora was already deleted by CivitAi.
@Elfenzorn I didn't download it in time.
Yes eventually. I have a few pipeline ideas, but will take a while. Curating video datasets is a pain in the ass and takes lot more time than just images.
@Elfenzorn can that be found anywhere else? Civarchive apparently isnt an archive because its deleted there as well.
@vackyadmiral167 Is it just obtaining the videos that is a problem? Because I may have a solution for you, or at least part of a solution.
@_GAi Not specifically just obtaining the videos but that could be a small issue. Need the videos to be precisely how you need them outputted during inference, have to find "clean" scenes where the desired behavior is there, finding the exact buckets where you have good behavior within 3 to 8 seconds of the clip etc.; then processing, captioning etc. But having good video sources always helps - and i have my own sources but happy to get your recommendations. Feel free to DM me to chat.
@_GAi yeah, MiniMax is great at sexy stuff and has lots of potential for BDSM, but the way it is out of the box it's quite limited and the loras are pretty slow in coming.
@vackyadmiral167 Awesome model. Thanks!
I'd been waiting for you to start applying your stuff to MiniMax. IMO it's a real groundbreaker of a model, but... trying to gen kinky stuff with it, it just needed a certain special little something to help it reach its full glory XDXD
Been using it for a bit now, while it's great it can also be a bit strange. The audio it generates doesn't seem to match the video very well.
I'm trying to get an ungagged character to make light grunting noises as the struggle, but its just defaulting to "Mmmh mmmh mmmh" on repeat.
Might be a side effect of using this with Ref2v or maybe something wrong with my prompt?
Possibly prompt because I’ve been able to fix a lot of problems by just more precise prompting, and/or strength too. It works fairly well around 0.7 which might help uncouple the speech patterns.
And if all that doesn’t work well chalk it up to my learning curve. It’s my first ever minimax Lora / video Lora with audio training.
@vackyadmiral167 Awesome model. Thanks!
I've been waiting for you to start applying your stuff to MiniMax. IMO it's a real groundbreaker of a model, but... trying to gen kinky stuff with it, it just needed a certain special little something to help it reach its full glory XDXD
Hilarious someone is trying to profit off the whip lora thay was recently deleted.
Ain't no one paying $10 for that
That ain’t me that’s for sure. I’ll host it for free on huggingface shortly.