>>281186738
>So the reason most AI clips are so short at the moment is because they are bottlenecked by VRAM?
In the case of wan2.2, yes. Everyone was hoping native 10 second videos would be part of this version but it wasn't, presumably because it's still hard for low-vram machines.

>That trick does sound like the way to make it longer, but i guess it would be more work
It's actually pretty easy. The issue is that each forthcoming gen can only work with the information it has from the input image.
In the case of this video, all information related to her face would be lost if you generated a video from the last frame. All the AI has to work with is an anime hand on a butt.