@Matt Sqwrl Hmm, yes, the image generators can be good for pre-vis type things, I would think, although image generators probably use a *lot* more resources than text generation. When I played with this stuff, though, I found that generated images can be good if it happens to hit on exactly what you want first time, but if there are mistakes (weird numbers of fingers, things not positioned as you wanted, etc) and you ask for corrections, each change typically throws something else off, then correcting that causes some other weirdness and you end up in a never-ending cycle of corrections. If you don't correct but instead edit the original prompt and start it from scratch, the more specific you make the prompt the more chances there seem to be for unintentional, non-obvious conflicts to cause it to glitch out. So yeah, quite apart from the ethical, environmental and resource-consumption problems, I really feel genAI is still at the stage of being an interesting research field, possibly good for playing around with, and, as you say, for speculative prototyping, but not suitable for real work. And yet people are relying on its unchecked output and even hooking it up to real world stuff with alarmingly gay abandon.