Content-Aware Fill analyzing takes too much time

[matheosct]

I want to fill the background of a composition with a picture in it but it takes too much time to AE analyze. I have a AMD Ryzen 5 2600 and a RTX 3060 while my pc at work is a GTX 550 and analyze faster than my home pc. What can I do to improve this process?

[Mylenium]

Without seeing the image and such we can't really tell you much, but based on your preview image using CAF on a black placeholder graphic probably doesn't make a lot of sense. The algorithm will simply analyze forever since there are no real patterns to match. That's not how this works and what it's meant for. You could cover up those holes with simple solids. Anything outside that will require more info.

Mylenium 

[matheosct]

It's not a black placeholder, it's a photo. And I can't use simples solids because it has shadows and textures.

[Mylenium]

Regardless, this is not an example I would consider particular suitable for CAF. You could still just duplicate the layer, move it into place and mask it with some feathered edges. I really think you're going about this the wrong way and use the wrong methodology.

Mylenium 

[matheosct]

Yes, I know, that was what I used to do. But the main point here is why my home pc can't stand analyzing the composition faster than my work pc since I have a better machine.

[Rick Gerard]

Content-Aware Fill is not the right tool for images. It needs footage, preferably with a frame-based production format, not an MP4. It is even better if you pre-compose any trimmed footage layer moving all attributes to the new comp and trimming the pre-comp to the layer length. You run CAF on the nested pre-comp in the main comp, not on the original footage. 

CAF also likes detail and does not respond well if there are not some closely adjacent frames that do not contain the object you are trying to replace. An aerial shot of a car moving down a highway is a good example. There are frames before and after the car moves past its position in the current frame that doesn't have a car in the way, so pixels can be chosen to fill the hole made by the mask. Without access to those pixels, you will have to create a PSD layer and create the fill manually. 

Processing speed depends on system resources, what is running in the background, and number of pixels you are trying to replace. 

[nubnubbud]

first off- welcome to "why video editors dislike vertical media",
regardless of how good your machine may be, AE is quite slow, and has very little hardware scaling. That means, so long as you meet hardware requirements, you're gonna see diminishing returns from better computer hardware.

AE is just bad on madern hardware- it was coded almost 30 years ago- before GPU's or multi-core CPU's, and you're telling it to use both. It simply cannot use GPU in most cases for anything, and only a few processes were ever made multithreaded, so it's definitely not going to help much. The only appreciable way to make it faster is to get a CPU with bigger, fewer, faster cores, and to just mash as much RAM as you can in a machine, which only helps you preview more than a few seconds at a time, and that's it.

As for your workflow- content-aware fill really is not very adaptive. it's buggy and AE's content-aware is different, as it needs to maintain consistency between frames- if you have only 1 frame, it doesn't have the data its code is asking for, and it won't do anything at all, like calculating X+12=?. there's just not enough input. 
Here's what you should do.
Either-
get a solid above and below, match them as close as possible to the edges' color. Add a gradient if needed. then feather the top and bottom of the image to blend it.
or
if it really is just an image, bring it into an image editor and extend the footage there.
or
if it's moving, consider feathering the edges, as above, but placing a larger, blurred video behind it. it would match up rather well in this case.