That's because a generative AI relies on learning through some datasets which generally includes images with perfectly visible faces and bodies but not fingers. In most images, human fingers are holding something or not visible at all, so the AI cannot model it perfectly.