I feel like it was inevitable that it would become slop. The models are impressive, but they can really only get you 80% there.

If you want a video of a dancing cat, sure, you can get that. But if you want an orange tabby doing the moonwalk or the robot, that's a lot harder. You'll have to generate dozens of videos and fine tune prompt incantations before you get what you want, if you even do before you hit a rate limit or you get frustrated. If you want something specific and unique and interesting, you still need to put in a lot of effort. Therefore, most videos that people actually make and share are pretty generic.

I think most art models have subtle tells and limitations similar to textual LLMs too, just a little harder to recognize. Certain ideas and imagery will be easier to generate and more likely to fill in the gaps of your prompt. The technology is fascinating compared to the nothing that we had before, but it still has real limitations - try to get it to generate an Italian plumber wearing a red hat that isn't Mario, for example.

All that to say, the trend towards low effort, repetitive, and uncreative results is inherent in the medium. Most users will prompt for a generic dancing cat and get something resembling a cat doing something that resembles a dance and that will flood social media. The few people going for a more creative and specific artistic view will be frustrated by the constant rolling of dice, and if they do make something they work hard on, it will be drowned out by the low effort slop posts. And if you're frustrated by those limitations and want to make something intentional, then you'll eventually gravitate towards Photoshop or Blender where you can actually craft the exact thing you want.

These models do not really "democratize art", they just make it really easy to generate visually interesting noise. Once the novelty wears off, the limitations are apparent. Art has always been democratized anyway - Blender and Krita are free, and pencils are cheap.