Trained on 140 images of diverse bondage positions and perspectives.
Description
FAQ
Comments (5)
I think a LoRA that has such complicated concept should weight more than 18MBs
That is not necessary true. If the model is working, is better to have a lower rank so it doesn't learn what it does not need to learn (faces, artifacts, etc)
This is the typical size when the network dimension is only 4. This is a indeed a too low value to understand many bondage positions and concepts. That's why the results are so bad.
If you want some real goods Loras about bondage, see those from Joschek. Each specialized Lora is about 400MB, but they're perfectly efficient.
This Lora doesn't seem particularly sophisticated to me: it completely changes the composition and very often produces mutated or absurd body shapes.
fair, its rather a small lora and in the end its still flux that doesnt work too well with nsfw