[...]
I tend to use the Low Resolution and Very Compressed options the most, though you need to compare models on each photo (you can view 4 results at a time).
Yes, a lot depends on the source image. I tend to take relatively Low-ISO shots, and then i can reduce the (auto-detected) 'image default' amount of noise reduction in GigaPixel to
very low values. This will make Gigapixel work even harder to find details. After all, some of the noise (shot-noise) is part of the recorded image itself. This also prevents a too 'plastic' look, and it minimizes hints of edge halo (if any).
Depending on the image, if there is mixed content such as vegetation and architecture, one can also create multiple upscales and then blend them with masking in Photoshop / Affinity Photo.
Cheers,
Bart