Raw LLM Responses

Disabled artist here. On the one hand, I do hate AI art for pretty much all the same reasons everyone else does. On the other … I do have to admit that it's useful for me, in particular, with my particular disability, if I can find an unlocked model that doesn't charge a subscription (because like most disabled people, I'm poor AF). See, I've got a traumatic brain injury, and one of the things I lost was the ability to do certain kinds of whole-image spatial manipulation in my head. As a result I have trouble with is layout and composition. When I "picture" an image in my head, and then try to draw it, I immediately discover that the image in my head was all weirdly non-linear and doesn't really fit on a 2d surface. If I try to draw without references, my people consistently get smaller from left to right—and if I stand back I can see that, but if I'm actually drawing it, the lines look parallel. Then there's all kinds of things where I can't "see" the composition in my head at all. I have huge problems with figuring out how angles of things go together. If I'm trying to draw a guy sitting at a desk, I can't "see" how those two things go together without having some sort of model in front of me. I can't "see" how to fit a scene into a frame. Now, for the final image, I solve this problem with 3d models. Some I make myself, some I get off Sketchfab, some I get off the CSP store. I'll meticulously set everything up, with attention to the parts I have trouble with like making sure the person's behind is on the chair, the chair is on the floor, the floor is flat, all that sort of thing. Then I'll take my camera and try out all kinds of angles, like I was doing an actual photoshoot. However, there's two problems. One, most artists "cheat" a little bit—they'll make bits bigger, smaller, off-kilter just a little, so that things look nicer. I can live without that, but it makes life hard. Two, and the bigger one, there's a lot of ways to lay out a scene, particularly when it's more conceptual. "A student doing homework in the year 2400" could look a lot of different ways, and I can picture exactly zero of them by myself. And this is what AI is *really really good at*. Even back when AI output was mostly abstract blobs, it was still better than I was at composition, because it had been fed a lot of layouts that humans thought looked good. It'll happily show me fifteen dozen ways to put a guy at a desk and make it kind of sci-fi-adjacent, and *get it all to fit in a frame*. I don't need to know what the desk looks like, what his chair looks like, where his hair goes—I'll be redoing all of that anyway. I need help with the big-picture stuff. And sure, I could go dig up my art books and trawl through pinterest or whatever, but the advantage of the AI stuff is that it's sort of the "normal" stuff. It's art regressed to the mean, which means that I can take that art and run with it my own direction, and not look like "Michael Whelan but steampunk" or "Frank Frazetta but pastel rainbows"—I look like I'm taking "normal" art and doing my own thing with it, which is exactly what I want. I want to *know* what normal looks like before I decide what to do, but I fundamentally cannot picture that anymore. So the AI helps. But also, like … I could get the same thing with AI models from like five, even ten years ago, so I don't really *need* these energy-wasting, job-stealing monstrosities. I just need *something* that I can actually get my hands on (which is a consistent problem: disabled people can't afford the subscription prices on any of this!) that can go a few feet past a google search for "stock image man working at desk" into "image of young man doing homework at futuristic desk in the year 2400" and output something where I can tell which part is supposed to be the young man and which part is the desk. And if you actually read all of that, here's a cookie: 🍪