MEGA does not follow my prompt - Maybe I found the problem/solution
Lets say you have a prompt like "the man pulls a handkerchief out of his pocket".
If you make a 5 seconds video out of it it seems like the AI does not follow your prompt and there is close to none motion going on at all.
But if you do the same prompt with a 2 second video there is way much motion going on an the AI follows your prompt way better.
It really seems like you need to make some prompts like "do that and than that and after that do this" for videos longer as 2 seconds. But not sure who do make the AI unterstsand those steps. Maybe someone can test this in more detail. I sadly do not have the time for it.
try wiring the Wan2.2 Pusa in (low noise version only). it makes things happen without having to script everything. Like people and stuff in the background doing things. At least for the V10 (and normal Wan of course) it helps a lot to avoid micromanaging everything via prompt.
Wan2.2 and the V10 version do actually work surprisingly with simple prompts. For example have simple prompt of three people doing something and drop it on a starting image with just one person. especialyl interesting in nsfw scenarios, It's quite good at filling the gaps itself.