I’ve been trying to curate a prompt for a few days now that can accurately recognize leg positioning on a quadruped to style or add “markings”, e.g., a custom-colored horse. It seems no matter how specific I am with the prompt (facing left, facing right, front fore leg nearest to viewer, etc.) the generated image is always off 99% of the time. Either it doesn’t recognize the leg positioning, it doesn’t draw the correct markings or it only draws 2 instead of 3/4. Even with multiple edit prompts, the image is still wrong.
Is this a bug or a limitation of the 4o image generator? I know the team said they were working on improving editing precision. I’m just not sure if this is something that will improve over time or if it’s fundamentally not possible with this model. I’d really love to be able to create accurately detailed characters 
