WAN 2.2 - Motion From Images

[RandomKaos999]

Hi all,

I'm wondering about training specific motion from images. From what I understand, and I admit my knowledge is limited. When you train WAN 2.2 on video input files, isn't it converted into images for the model along with caption(s) in order for it to learn a specific motion ? I'm just wondering if its possible to entirely skip the step of providing video input, and if its possible to provide mutiple image sequences which shows the motion wanted to train?
Has anyone tried this? - I may be well off base with my understanding, wouldn't be the first time, but I would image if each video frame is extracted then it would be possible to train motion based on pure images only?
If you have attempted this, can you let me know how it went on please. I've tried a couple of things but not really got anywhere and just wondering if its actually possible or not. Albeit, I trained with a single image sequence so it effectively learnt the entire contents of it - but im just wondering if its possible to provide multiple sequences to create a motion lora and how to effectively do that.