Prompt Engineering vs Structured Prompts

I’m experimenting with using SML (a small markup language) to describe scenes
instead of writing free-text prompts.

Example:

Scene {
    character: "man", "woman"
    pose: "yab-yum"

    Camera {
        angle: "front"
        light: "sunset"
    }
}

Grok actually generated the pose more accurately than when I tried
prompt engineering.

Video of the experiment:
https://www.youtube.com/watch?v=vRAPoGxnNwk

I'm curious:
Have you tried structured prompts instead of prompt engineering?