This is a workflow for outpainting videos.
It creates a padding around an existing videos and expands it.
This workflow also comes pre installed with my Wan RunPod template:
https://runpod.io/console/deploy?template=758dsjwiqz&ref=uyjfcrgy
You can watch a short tutorial here
Description
FAQ
Comments (8)
Native wan inpaint has almost no trouble doing that either, simply using kijai's padding with colour=0,0,0, and prompting "any dark hidden regions instantly vanish" (white may be a better choice).
Good to know there are more approaches to this
Yeah, I'm not too keen on the overhead of model switching (at least afaik vace still hasn't released 14b? might be another week), though vace seems to do a much cleaner job of it.
Might even be better not to refer to it by colour since it sometimes causes corruption, now just "any hidden regions unrelated to the image are revealed" or something. I've also switched to 0,256,0 and greenscreen in the negative (hoping the model knows that keyword), seems to perform better so far.
For "dark hidden regions" are you saying if the input image has for example black bars on the sides?
@Kiefstorm outpainting would imply there's some form of empty region around the input image. So the original input wouldn't but the original + the new sized output would
@Kiefstorm Precisely, like those all too common vertical shots, or add them yourself. And not just on the sides - you could crop out areas within the image to fill those too. Simple padding is restricted to the edges though.
Does this thing know what a Lora is? :D
I couldn't figure out how to hook a LoRa up, if anyone knows I'd appreciate. thx