A "point spread function" denotes how a single point from the source, or a punctual object (mathematically a Dirac) would spread on the observed image through the imaging system.
In blind deconvolution, both image priors and PSF priors are useful to unmix those intricate systems (even more with noise). They are different, as the PSF is somehow invariant to the imaging system (there exist space variant PSFs or blurs, let us skip that for the moment), while different images can be acquired.
PSF priors may be related to some parametric shape (e.g. a Gaussian of unknown width), a bounded support in space or frequency, smoothness, etc, while data priors are often more generic to some image class, related to sparsity, bounded variation, etc.