afaic prompting is like tea leaf reading, you can write six ways from sunday "blue shirt" "(blue shirt)" "((blue shirt))" "((blue shirt:1.6))" and the sd model will be like "ah yeah red shirt coming right up fam!"

Mostly, yeah, although some of that is UI frontend formatting. For certain frontends and model types something like "(tag:2)" increases the weight of the tag during the turning the text into usable numbers stage, and it only did anything if that was actually a tag the model or lora was trained with. It had some limited ability to force like SD1.5 or SDXL based models more or less towards a concept, but there's always so much random noise and incoherence that means actually making the shitty gacha churn out a desired result means lots and lots of rerolling and poking at the prompt and it never actually does a good job.

Modern qwen based natural language prompt models are literally just you describe something in as much detail as possible and then the image model gives something that's still dogshit and still randomly broken, but is a little more like what it's told than the older ones did.

There's no secret to it, and even at it's most esoteric it was less complicated than the markup formatting used in reddit or lemmy posts lmao.