In your denouncement of the ostensibly trite content of "a t. rex bursting out the center of mars surrounded by rainbow mandala constellations or whatever", you fail to look past the content itself and find value in the technique. From an artist's perspective, these models are nothing short of magical in how they manage to pack in knowledge across so many styles and contexts.
It's personally mind-blowing how effective diffusion is in general (but retrospectively, it's a beautifully simple concept), but it's not just the isolated image of a t-rex on mars that it interesting. It's the speed and cost at which the image can be generated, and the ways in which the data can be modified or recontextualized in a feedback loop. Or how multimodal systems like GPTV or LLaVA allow a new form of communication which blurs the line between images and language. I could go on, but the essence is that you are doing yourself a severe disservice by generalizing these capabilities as "blockbuster clip art".
It's personally mind-blowing how effective diffusion is in general (but retrospectively, it's a beautifully simple concept), but it's not just the isolated image of a t-rex on mars that it interesting. It's the speed and cost at which the image can be generated, and the ways in which the data can be modified or recontextualized in a feedback loop. Or how multimodal systems like GPTV or LLaVA allow a new form of communication which blurs the line between images and language. I could go on, but the essence is that you are doing yourself a severe disservice by generalizing these capabilities as "blockbuster clip art".