Question about random sampling

[LeiLiu-cloud]

Hi,
Thanks for your repository and it helps!

I recently trained the rock image and generated random samples with injecting scale =0. It indeed generates different rock images but is very coarse (I wonder if it is because scale 0 is the coarsest scale?). If I directly inject the image to the fine scale (for example 6), the generated image is almost the same as the original with no variations at all.
I'm not sure if I misunderstood the random sampling process. Should we start with scale 0 and then inject the generated image at scale 0 to the fine-scale model?

Thank you for your time!