>>3004034
I am still figuring this shit out myself out. Right now I'm testing two options:

1. Image to Video, with first frame predetermined.
2. Image to Video with first and last frame predetermined.

Option 1 is easier since you only need first image, but video tends to break apart towards the end,

Option 2 get way better results but obtaining two images that work well together is pain in the ass.

>>3004028 this video uses Option 1. You can try the workflow (https://files.catbox.moe/fmaeg8.png) by pulling it to the Comfy and installing the nodes.

<This video technically uses Option 2 with first and last frame using the same image to enforce looping. Workflow here: https://files.catbox.moe/r4g0xt.png

Like you said, Comfy is suck at being functional so it might take some doing to get these workflows working.

I don't want this comment to be any longer so here is links to models that I use. Good luck!

https://pastebin.com/vbipWi4w