Since PSFEx generates a model using basis vectors, with the linear combination dependent on image position, this is used to collapse that into a useable kernel for convolution and insertion purposes. See https://psfex.readthedocs.io/en/latest/Appendices.html This function will return the PSFEx output to the pixel scale of the original image. To keep an oversampled PSF model image, use load_psfex_oversampled() :param model_path: Path to PSFEx-generated model, as a FITS file (usually ends in .psf) :param x: pixel x-coordinate to use for model input :param y: pixel y-coordinate to use for model input :return: numpy.ndarray representing the PSF model as an image.