Feeding and learning AI

it would be nice, if the AI could learn, like: Generate a composition of two objects "A" and "B". "A" is the letter V made of black feather, "B" is  the letter S made of white feathers. Place "A" behind "B". The more prompts i use, i get coplete diffrent results. the prompts will only be partially implemented if there are too many (or if it is not too general / or if the description is more detailed). I cant get a V from black and an S from white feathers, but get a VS from mixed black and white feathers e.g.