I’m not up to speed with the whole post, OR the water system in .26.
however, having implemented vectrofields across anything in my project last month.
with that, im rendering a localized texture from an overhead camera that catches only the elements needed in one pass, and subsequently packs a render of the scene depth into the B channel.
If you make something similar you can just promote whatever texture sample is in use to a parameter, and swap it out for your own custom texture in the material instance.
Just make sure you aren’t rendering stuff you don’t need.
even limit to just the landscape. And use a post process material to get the depth output into B.
(0,0) append depth.
many ways to write different captures to the same RT - but if you do just 1 pass its less expensive.
/ I would imagine that if they had implemented it they would be using something pretty similar for localization.
in the video the texture seems rendered just to experiment.