AI still hates us, again.

More about AI still hates us, again.

I was at my energy limit anyway, so figured I might as well give it a shot. Just as I feared, making slight changes to that part of the prompt and emphasizing them still doesn't get the red stripes back. And otherwise there's problems, the AI inexplicably ignored the request to have her walk to the left that worked fine in the previous attempt, instead walking to the right and showing objects and details that were not there the last time the camera gave us a view of that area. It appears to have seized on part of the prompt described as seen through the doorway in the new room to completely reimagine the already seen area, so suddenly there's a bed smack dab in the middle of the room, and curtains on the window.

That's my biggest complaint about the whole concept of AI images and video. Every AI platform I've experimented is HORRIBLE at following directions. Not to mention wildly inconsistent at producing even broadly similar results from the same prompting. Sure, this is a beautiful clip in its own right, but completely unusable because of all the glaring continuity errors. Very sad, because from my limited explorations DDG's ProVideo is capable of some very impressive video making, just needs more ways to nail down all the details to keep the clips plausible between takes.

Some actual documentation for prompt engineering would not hurt in the least, either. Anything generated in a previous clip should not be such a huge struggle to produce again in a subsequent generation. I have probably a half dozen clips produced in this campaign which show the detail I want, so I *know* the model can generate that detail. But, alas, the mindless jerks at the Sirius Cybernetics Corporation do not believe in manuals. "Glad to be of service." :eyeroll