Selectively grab regions of pixels from one image to transcribe to another

I want to take parts of an image (img1) and place them into a new image (img2).

What I am doing is selecting specific ranges of pixels from img1 and placing them in the same exact place as img2.

So what I did was:

     img1 = I{1}; %just a cell with several images of size (256,256)
     img2 = zeros(256,256);
     img2(1,1,46,46) = imcrop(img1, [1,1,45,45]); %These pixel values are simplified for the example

I also want to do this iteratively, meaning that I'm adding several of these "squares" randomly taken from img1 into img 2.

In simple code it might look like:

    for square = 1:50:300
       img2(1+square,1+square,46,46) = imcrop(img1, [1+square,1+square,45,45]); 
       ...
    end

The code is a touch more complex than this (especially picking the squares) but the real issue is that the img1 doesn't transcribe well into the new img2. I'm getting a white image with some dark spots but not the images that I would like to see.

Here are some examples:

and an example starting image is:

Note that the starting image wasn't red (grayscale), this example image already had some processing done on it.

If I don't declare img2 = zeros(256,256) I do get the squares that I want but for some reason they are all disorganized and not usable for my purpose. Here is an example of that same image without a declared img2:

I'm not really sure what Matlab is doing so any advice would be great. I think that imoverlay might be a possibility but I'm not 100% if that is the best way to approach this problem.