The AI will always struggle for the same reason the AI does in every category. People assume its magical, or aren't paid enough to give a shit, so they do zero vetting or editing on it once its done to make sure the parts where it failed to understand the intention are corrected.
It will still need multiple takes to get some scenes right, same as a human, simply because it needs directing on the correct emotion and all that jazz. But as we can already see, companies aren't going to do that so the final product always ends up slop trash.
Now, unofficial works will likely be much improved to the point of superior because passion projects will have guys willing to do that work. Those are the guys that make the AI art right now that is indistinguishable because they've spent hours messing with the prompts and rejecting dozens of lesser versions before considering it done.
In theory you should have one person overlooking to make sure it still works, but we all know these companies are too cheap to do that so it will take a catastrophic failure to get them to hire an overseer.
Probably should have a few really, both in terms of effeciency and extra sets of eyes, relative to the size of the project, but what keeps happening is one guy does literally everything using a prompt/program left over from another project.
It is continually amusing that the same problem you might see on a lazy MTL scanlator working by himself on a niche series repeats with billion dollar corps working on major properties.
The AI will always struggle for the same reason the AI does in every category. People assume its magical, or aren't paid enough to give a shit, so they do zero vetting or editing on it once its done to make sure the parts where it failed to understand the intention are corrected.
It will still need multiple takes to get some scenes right, same as a human, simply because it needs directing on the correct emotion and all that jazz. But as we can already see, companies aren't going to do that so the final product always ends up slop trash.
Now, unofficial works will likely be much improved to the point of superior because passion projects will have guys willing to do that work. Those are the guys that make the AI art right now that is indistinguishable because they've spent hours messing with the prompts and rejecting dozens of lesser versions before considering it done.
In theory you should have one person overlooking to make sure it still works, but we all know these companies are too cheap to do that so it will take a catastrophic failure to get them to hire an overseer.
Probably should have a few really, both in terms of effeciency and extra sets of eyes, relative to the size of the project, but what keeps happening is one guy does literally everything using a prompt/program left over from another project.
It is continually amusing that the same problem you might see on a lazy MTL scanlator working by himself on a niche series repeats with billion dollar corps working on major properties.