This model was trained on a dataset of 2,000 images covering various N5fw and BD5M concepts. It focuses on reference-based image editing. For optimal results, use a subject reference, followed by references for the type of gag (ball gag, ring gag, etc.) or restraints (rope, wool, arm binder, box binder, cuffs, etc.), as well as other additions (posture collar, nose hook, blindfold, nipple clamps, etc.). For Tensor.art, if a prompt is too specific or gets censored, try adding spaces or line breaks. On TensorHub, you have more freedom with prompts, without censorship.