Making something bigger

One kind of upscaler enlarges what it was given and works hard not to change it. Another adds detail that was never in the source, which looks better on a soft photograph and is dishonest on a product shot.

Which you want depends on what the picture is for. For a catalogue, hold the detail. For a background plate nobody will inspect, inventing is fine.

Some upscalers charge by output size, so a large enlargement of a large source costs a lot more than a small one. At least one here charges a flat amount per image whatever goes in, which makes a batch predictable before you start it.

Both are on the button before you run anything. The flat one is usually the right default when you are processing a set.

Upscale last. Editing re-renders the picture, so anything you gained by enlarging first is spent again, and you pay for the larger frames twice.

A video upscaler has an extra problem an image upscaler does not: the frames have to agree with each other. Upscaling each frame independently gives a clip that shimmers, because every frame invents its detail differently.

That is what temporal consistency means on those model pages, and it is the reason not to run an image tool over a clip.