documentation suggests they can (says "imgs: 3D/4D Niimg-like object").
example:

from nilearn import datasets
from nilearn.input_data import NiftiMapsMasker, NiftiLabelsMasker


img = datasets.load_mni152_template()

atlas_img = datasets.fetch_atlas_destrieux_2009()["maps"]
labels_masker = NiftiLabelsMasker(atlas_img)

prob_atlas_img = datasets.fetch_atlas_harvard_oxford("cort-prob

This line https://github.com/SIMEXP/niak/blob/master/bricks/fmri_preprocess/niak_brick_regress_confounds.m#L234

should say "Adding motion parameters..."

and not "Adding high frequency noise..."