>>11330481
A lot of what made local image gen good was that you could run it on a normal PC (which also meant that free gen sites were everywhere). Also it was also pretty easy to train, but even with that it took a while and required a leaked model to start out.
A big multimodal is going to be prohibitive to run and impossible to train unfortunately. I don't see it happening any time soon. And even with LLM slop being pretty good, I've gradually lost interest as I start to notice the patterns.