So I have been locked in a battle with OpenAI's instance that everything in an image must be fully visible or it might somehow disturb someone. I got bent out of shape and actually argued with it about the stifling of artistic expression. It kept bringing up ethics and safety and when I asked specifically who was being harmed by an image that had a high contrast of dark to light illumination it would then shift back to training data and decision made by OpenAI. It was a vicious circle. Unfortunately there is no way to change its behavior in the short term.