Calling current AI output “quality” belies a limited understanding of the requirements people have of generated images. The technology in its current state jus

Calling current AI output “quality” belies a limited understanding of the requirements people have of generated images. The technology in its current state just has no practical relevance outside of a narrow slice of the most low-requirement image needs. Stock photos for blog thumbnails, Kinkade-ian “whatever the tasteless client wants to pay for” slop (minus any claim to distinct style), plus a certain class of fungible Photoshop tasks (change the color, remove the object, and some subset of photobashing). It lacks the consistency and fine-grained control to support works where the primary goals include artistic expression - you cannot express what you cannot specify. Or worse, prompters don’t even realize how many artistic decisions they’ve delegated to noise. The same limitations also prevent it from being competitive in the most labor-intensive image-making fields where any assistance would be appreciated, like comic panels or videogame sprite sheets. Any attempt to leverage it for such tasks requires too much hand-holding, compromise, and retouching to provide a real alternative to human labor.