Works well with base and distilled models both. Use strength of 0.8 to 1.2 - 1.0 works best on average, but if your image is slightly lower quality then reduce strength.

Follow the trigger framework for best consistency; natural language also works but needs the anchor elements (the first two bullets below):

• Prompt anchor: "A woman in elbow-bondage pose with her arms tied behind her back."

• Then add: "[color] [materials: ropes/ties/straps/tape] wrap around her wrists and elbows."

• View: you can specify where her torso is facing [rear/sideways/front] and the direction of her head whether looking away, somewhere else, or at the viewer.

  • Front pose is consistently the worst. Needs a bit of an angle to ensure its showing off the tied elbows right.

Provide feedback if its breaking somewhere.