Order of operations matters more than model choice when upscaling old footage

A denoise pass and an upscale pass are not commutative. Doing them in the wrong order
permanently bakes artefacts into the result, and no later step can undo it.

The wrong order (what most people do)

upscale 4x  ->  looks soft  ->  sharpen  ->  looks crunchy  ->  denoise

By the time you denoise, the noise has already been magnified and reinterpreted by the
upscaler as if it were real detail. The model happily rendered it as texture. Now you are
trying to remove structure that the model deliberately invented.

The right order

denoise  ->  stabilise  ->  colour/levels  ->  upscale  ->  light sharpen

1. Denoise on the original resolution, where noise is small and separable.
2. Stabilise if handheld — motion blur gets amplified just like noise.
3. Colour and levels — fixes are cheaper before the file is 4x bigger.
4. Upscale last, on the cleanest possible input.
5. Sharpen only lightly, and only if the upscaler left things soft.

Why this matters more than the model

Two different upscalers fed the same clean source differ by maybe 10 percent in
perceived quality. The same upscaler fed a noisy versus a clean source differs by a
factor of two. Input hygiene dominates model selection.

A tool does not fix a workflow. UpgoLAB will upscale whatever you give
it — including a noisy source, faithfully magnifying the noise. Feed it the good version.