I am not an ad person, but to me they always contain a specific subject (thing, place, etc). If you are making an ad for widget XJ900, you don't want it to actu

I am not an ad person, but to me they always contain a specific subject (thing, place, etc). If you are making an ad for widget XJ900, you don't want it to actually show the XK800. What's harder, trying to prompt the AI into generating content on your specific subject, or just modeling/filming the subject with current methods? Which is harder to incorporate client feedback? So far, slight tweaks in prompts can produce radically difficult results. Paint back can help, but we'll see. I have yet to see a stable diffusion output that is high quality work. They often lack any compositional sense, and all the lines are weirdly wavy. It's like someone trying to do realism but with the motor skills and spatial awareness of a toddler. The picture that won the state fair awards looked really terrible on further inspection. Basic composition elements like the horizon line or the giant circle in the middle were all skewed, and not in an attractive way. Any elements beyond the central figures are just noise. I see that story as more about bad judging than anything else. Stable diffusion has its own version of uncanny valley, and it may be a long time before we can cross it. The people who decree stable diffusion as the death of human art don't often to have a firm grasp on artistic or design fundamentals.