So, I don't call it "Generative AI" because Intelligence really isn't involved in current systems that are creating funny pictures of Obama dabbing infront of a burning Reichstag. The capacity of algorithmic decision trees and weighted neural networks is impressive, but that human operator element is required to steer the software into an acceptable result. What has been happening is not that operator inputs are less needed, but rather, the threshold of understanding operators need to have for the technology is getting lower.
Short comment, but it's long enough to need the read-more.
Part of this is due to the continued improvement of various text-based input methodologies, through improving their specificity and through adding additional manually curated databases to accomplish specific tasks (eg: LORA's). A big step recently is that MidJourney can now utilize a single reference image of a [subject] and then recreate that [subject] in novel poses with relatively good effect on the detailing. Depending on how quality your information provided is overall, and the data you have referenced, you might see very soon that indeed the way we used to think about MAGI as merely copying information will become moot.
New programs out there use decreasing amounts of direct visual reference, often small quantities of basic visual references are sufficient to "train" these newer modalities, some of which are built on old databanks, and others are less so. For artist-types (myself included in those ranks), I think there is about 2 years of time tops before professional illustration requires utilization of these tools as part of your training and education. Adobe has plug-ins pushed into Photoshop and their other creative suite products, and if you're not watching the space you will be startled to find out how much more accessible these programs are becoming.
What we're likely to see here is the adjustment of software into 'creative suite' style interfaces, primitive GUI's like NMKD won't be the standard for much longer. Control Maps and that aforementioned single-reference system are going to make it such that in software, you could pose characters based on a reference and then generate landscapes with likely in-software tools. Adobe's products are starting to resemble this, and I don't think there's a way to close this particular box. I'm not jazzed about it, before you think this is a glow-up, I'm looking at the devaluation of human labor and seeing the writing on the wall.
If you're an artist type who has enjoyed simply posting into the internet without putting yourself out as a 'personal brand' in the Parasocial manner, you're going to struggle more and more against this technology. If you're not making physical goods, you especially are not going to get far without having some personal brand or manage a unique style. It feels like once the new built-for-purpose computer chips that are made for these systems start to arise, we're going to see the trend escalate.