This depends on the format of your input data, and if you can use an 'Easy Mode' like qiime rescript get-silva-data.
The full pipeline is described in Getting SILVA data: Hard Mode. You import the three input files, then reverse-transcribe and parse-silva-taxonomy.
Each file imported has a --type, like --type 'FeatureData[SILVATaxonomy]'. Those types are listed here. The input test data for the RESCRIPt plugin can be used as examples of that format.
At the risk of stating the obvious, this process is hard and technical. I try to avoid this by using prebuilt classifiers, or databases already in the Qiime2 format.
What database are you trying to import?