This is an early test version of the model. I trained it fairly quickly using a limited dataset, mainly to understand how the model responds and behaves.
The dataset contained only 15 images and was trained for approximately 3,000 steps. Its general trigger words include terms such as classic BDSM, ropes, tied, and similar concepts.
For the best results, use detailed prompts that clearly describe the image, the character, her appearance, pose, expression, and positioning within the scene. Being precise about the desired composition and action will help guide the model more effectively.
The model performs well at strengths between 0.5 and 1.2, although I recommend staying between 0.5 and 0.8 to preserve image quality and maintain the natural appearance of this beautiful model. steps from 6-12 depending.
Regarding samplers and schedulers, Euler and Euler A seem to produce particularly good results. RES2, ER-SDE, Simple, Beta, and most other combinations also work well. The model is fairly flexible, although you may occasionally need to reinforce certain details more strongly in the prompt to get the exact result you are looking for.
This release is only an initial test, and an improved and more thoroughly trained version will be uploaded soon.
feel free to swing by if you have questions. I'll update soon anyway.
https://t.me/+BrNISE72oLYwNDMx
https://t.me/ccmb1ne
Description
FAQ
Comments (2)
This works pretty well, it seems to strongly prefer wet skin even when prompting against it when the conditioning node or the bypass lora are enabled.
Without those, the ropes typically aren't rendered. Still this is pretty good as is, can't wait to see what else you cook up. Great work!
This is a problem with Krea2 in general in my experience. Wetness slider LoRA https://civitai.red/models/2738333/wetness-slider-krea2-zit?modelVersionId=3079282 with -1 weight helps a lot for my cases.