BRKN-BDSM This is an highly experimental Lora , I tested something with an eclectic dataset. It respond very well to intense emotion and different style but lack consistency for now . This is the V1 I will had to it and correct some weakness of the Lora as I go along . I trained it over 5k for about 115 images in BF16 . For this slow LoRA, instead of training on a single specific act repeated over and over, I built it around a broader practice and spread it across many different forms, techniques, and environments. That does introduce some “pollution” you’ll see things like hooks, Hitachi-style wands, and other gear mixed in but it also gives the model room to surface interesting variations instead of being locked to one exact object or move.I focused primarily on the bondage side: all kinds of restraints like rope and shibari work, metal cuffs, and leather straps. I didn’t really dive into the gag territory (spider gags, ball gags, etc.); the emphasis was more on the movement and dynamic of the action than on specific props, outfits, or styling. I’ll be pairing it with a dedicated prompter tuned for Z (but usable with Flux and others) that I’ll share today, and that should make prompting this LoRA especially fun and effective. This LoRA also works across a range of settings, from classic indoor spaces like dungeons or cave-like chambers to eerie outdoor environments with a “creepy woods” vibe. I didn’t lean heavily into industrial locations, but there’s a touch of that aesthetic in there as well.The dataset is built entirely around restraint suspended, or not, shibari, but not reserved to it, and this is why I didn't push it in the triggers. All in multiple poses – lots of open, stretched, contorted positions, from “pretzel” shapes to more elegant, display-style poses like a twisted secretary stance. At its core, though, the focus is on emotion rather than just the positions themselves, and that’s where Z really shines as a base model. This LoRA is designed to take full advantage of Z’s strength in conveying mood and feeling.

Files with SHA256: 8686f85b3f55bc40461fe17af56746550bcc1ed9d817e19040ae5e6e36e905ec