Why is it so hard to generate images based on a description of a simple sex position, like when the woman is lying on the bottom and the man is on top of her? The man or woman always ends up positioned incorrectly.

Here is an example prompt: A photorealistic hentai marker-style ultra-high definition frame depicts a top-down view of a luxurious candy-pink maximalist bedroom bathed in golden hour lighting, casting long soft shadows and saturating the scene in warm pastel tones. The composition centers on a full-body shot of mature courtesan Tynnyfer Johansen, her Romantic Wave blonde hair spilling across the pillow as she lies beneath a tall black man in missionary position penetration. His muscular back dominates the frame, completely covering her hourglass figure—her goddess-like breasts pressed against his chest, her thin waist accentuated by gripping thighs wrapped around his hips. His face buries into the pillow to her right while her deep blue eyes, framed by visible age lines, flutter with pleasure beneath him. Their naked bodies glisten with subtle sweat under the directional sunrise light, every anatomical detail rendered with Paul Lung’s precision—from the tension in his thrusting hips to the coy flirtation of her parted lips. The scene vibrates with Queen Bee-style cinematic energy: garter stockings stretched taut around her calves, plastic-perfect pink satin sheets rumpled beneath them, fourth-wall-breaking intimacy captured through Toshio Maeda’s signature expressive linework in the man’s straining tendons and her clutching fingertips.

And the closest thing I managed to generate using SDXL.