Hi @Matilda_H-D,
Thanks for posting your question! The metadata input to filter-seqs can only consist of feature metadata, e.g., a sequence file or taxonomy file. We do not yet have functionality for removing sequences that are found in a specific sample in a single command (multi-step details are below).
We plan to add tutorials describing this process in this month's release of QIIME 2 so check back then for more details.
For now, this forum post still describes the best workaround for removing features from a feature table that are detected in a specific sample. What filter-features allows us to do is to also filter our sequences file by passing in features-to-filter.tsv (see the forum post for how that file is generated) as metadata.
Note, however, that removing all sequences found in a blank may not be a good approach; many of these sequences may in fact be cross-contaminants rather than exogenous contaminants and removing them could eliminate valid features from other samples.
In the future we plan to add methods for contaminant detection that more directly address this issue.
I hope that answers your questions! Please let us know if you have additional questions/concerns.