I think prompt adherence is very important. You can modify and adapt styles, but getting exactly what you want in complex prompts is a really difficult task at the moment.
Definitely hard disagree with something like this as ai is easily able to recreate realistic images because those things already exists it's already in its memory it already knows how to do that whereas interpretive or surrealist images are harder for AI to reproduce because it doesn't actually exist in its memory AI cannot actually create things it's never seen before as much as is able to combine elements of existing memories. Therefore being able to adapt to unforeseen situations or surrealist images is objectively more impressive
@@GumOnTheWall Can't argue with that. For me, however, I wan't to tell stories and make movies. In doing so, I need imagery of basic things like two people sitting at a table and talking, arguing, walking into a room, running. I have a hard time getting AI to generate a basic "over the shoulder" shot. I've had no problem getting AI to generate surreal imagery. I tends to want to do that more often than naught.
@@TheinterfaceTvSeries yeah I think that's not an issue with understanding realism that's an issue with understanding language AI is not very good with perspective or points of view but I think stable diffusion 3.0 is supposed to be the best at understanding perspective so I think with SD 3.0 it might be a thing of the past
Not keen on talking about models that are not out yet. The images could be handpicked by SD. Out of 100 generations, you will get one with the right spelling, or correct position.
But still SD 3 will be as good as dalle 3 if not more, and considering it will be an open source model, it will definitely have less restrictions and censorship than dalle 3.