Hi @Alan_Chan,
Normalization methods like CSS or DESeq are not available in qiime2 yet. The choice of doing these outside of QIIME2 and re-importing is totally up to you and your specific questions in mind. Depending on what you plan on doing you might not have to do those at all.
For example, ANCOM (and gneiss) don’t require any normalization as they use log ratios of your composition data (pro!), though they do have certain assumptions of their own which you’ll need to account for (con?). You wouldn’t have to leave qiime2 (pro) but the current version of ANCOM in qiime2 doesn’t allow for longitudinal comparisons and mixed effects models (con) though you can with gneiss, see tutorial here. The taxonomic analysis in the moving pictures tutorial simply provides barplots of your taxonomic composition but doesn’t perform any statistical testing. The barplots are more for exploration purposes. With ANCOM you can actually test for differential abundance of individual taxa across groups.