I generate the images locally on my machine using ComfyUI, then it's just a mixture of tagging for quality (I use embeds so I don't have to constantly retype things like masterpiece or whatever), upscaling the latent, running outputs through a second sampler (not everyone does this and I don't even know if I should be doing it), then resizing it back down.
So for instance, the outputs are actually generated at 1024 x 576, upscaled 3x to 3072 x 1728, then resized to 1920 x 1080.