Movie Render Queue: Get similiar low resolution and high resolution images


I am trying to generate a dataset with Unreal Engine 5.4.
For my dataset I want a 1920x1080 (Full HD) image and a matching 3840x2160 (4k) image, to use it and train a super resolution algorithm.
The FullHD image should have no anti aliasing applied to it, the 4k image should have anti aliasing applied to it.
To generate my dataset I am using the Movie Render Queue plugin and enable the anti aliasing option. For the FullHD image I set the option to None and Spatial and Temporal Sample to 1.
This leads to the desired outcome of having aliasing artefacts all over my image.
For the 4k image I set the Anti Aliasing option to None as well but the Spatial and or Temporal Sample to 8. This leads to the desired outcome of having anti aliasing applied to the image.
(Notice setting Anti Aliasing to None is recommend in the Documentation of the movie render queue, as the multiple spatial/temporal samples lead to aliase free images!)
The only problem is that my images have lighting artefacts. which in some cases are not noticable, but in other they are severe.

Here the FullHD image:

Here the 4k image but with spatial and temporal sample to 1 (which leads to the same lighting)

Here the 4k image with spatial sample set to 8:

Here the 4k iamge with temporal sample set to 8:

And at last the 4k image with spatial and temporal sample set to 8:

Notice that all of the spatial/temproal sample images have different lighting (most noticable in the top left corner).

As I want my network to “only” learn how to increase the resolution and apply anti aliasing, I want the images for my training to be as close as possible wih as little lighting differences as possible.

I do not understand the rendering pipeline of Unreal, therefore please explain to me if this is the expected behaviour of my usecase.
Any recommendation of how to generate my dataset would be very welcome!

Thank you very much for reading to this point!