What makes you think that AI art generators getting "better" (for whatever definition of better you see fit) will make them less unpopular? They're not unpopular because they give poor quality output.
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.
IMO the important sense of "better" will be good integrations - making various ML-based tools convenient and common to use as part of a workflow.
We're all biased towards morally justifying what we're already doing. Those against AI still tend to use tools like Google Translate (trained on translators' work, partially displacing their jobs) and when asked will often give reasons why it's different and other AI tools are worse; ultimately I think that's ad-hoc rationale downstream from the fact they were already using Google Translate (because it's useful and baked into their browser) rather than it being reasoning that they actually used to determine whether to use Google Translate and which would need addressing for other ML tools to become accepted.
If someone starts using, say initially, ML-based background removal and upscaling as part of their workflow (because it's demonstrably useful and easily accessible in their editor), I think they'd equally start justifying that.