Hi again,
I would suggest running the rarefaction plot without using metadata as one of the inputs. Otherwise, you will get curves that represents groups of your metadata. And such curves are not expected to level out exponentially (they encompass sets of samples with different diversity) as seen for per-sample rarefaction curves.
The bottom plot represents the number of samples in each group that has at least that number of reads (x-axis). For example, we see that all 4 of your "blue group" samples have at least 500 reads, but only 2 out of 4 have at least 600 reads. The bottom plots in the tutorials probably were not using metadata so that each of the bottom plot "groups" had a single sample with the "groups" "defined" by the sampleid, which results in several lines overlayed at y=1 (the number of samples in each "sampleid group").
Sounds confusing, but hope this helps