The failure
We wrote five plus size casting briefs. Each one said the model was genuinely plus size, around a UK 20. Each one said, in as many words, do not render a slim woman. All five came back as straight size women with slightly fuller faces.
The obvious response is to say it louder. We tried that. Capitals, repetition, an instruction to read the rule twice. None of it moved the output, which is the useful part of the story: the problem was never weak emphasis.
What was actually happening
Further down the same prompt sat a line describing the model's presence. It read, roughly: professional editorial fashion model, strikingly attractive, fine symmetrical features, elegant bone structure, poised composed carriage.
Read that as an image model does. Almost every photograph it has ever seen captioned that way was a straight size woman. The line is a description of a face, but it carries a body along with it, and it was the last thing said before the picture was drawn. Given two instructions pulling in opposite directions, the model resolved toward the more familiar image, every single time.
Three changes, all of them needed
First, the body rule moved to the end. Whatever comes last does not have to argue with anything after it, and in a long prompt that turns out to matter more than any amount of capitalisation.
Second, the presence line became build aware. A plus size model no longer gets elegant bone structure. She gets a sentence saying she is beautiful and she is large, both fully true at the same time, so the two halves of the brief stop fighting each other.
Third, the body rule overrides by name. Anywhere above this line calls her slender, elegant, editorial, fine boned or poised, that does not apply to her body. Naming the contradiction is what defuses it. A rule that simply asserts something is a vote; a rule that overrides is a decision.
Geometry beat adjectives
The wording that finally worked contained almost no adjectives. Her upper arms are thick and clearly wider than her forearms. Her stomach is round and pushes the fabric out in front of her. Her bust is heavy and low. There is a soft double chin.
And then an exit test, which is the cheapest and most effective line in the whole prompt: if the woman you are about to draw could model straight sizes, start again.
A second, separate problem
Even with all of that, the casting portraits still read as straight size, because they were framed waist up. A crop above the hips removes the evidence. No instruction can restore information the frame does not contain.
Curve and big and tall portraits are now three quarter length with the arms held away from the body. That change alone did as much work as any sentence.
The general lesson
When a prompt loses to a prior, the fix is usually structural rather than emphatic. Move the rule to where nothing can contradict it, name the thing it overrides, describe geometry rather than qualities, and check that the frame can even show what you are asking for.
Age turned out to need the same treatment for the opposite reason: ask for a woman in her sixties and you get a thirty year old with grey hair unless the correction is explicit and carries a floor. Kidswear did not, because nothing else in that prompt was arguing with it.
