Noise ceiling on fMRI signals without repeats

I am trying to calculate noise ceiling of the fMRI data collected when subjects are watching movie stimuli. Usually, fMRI data collection for the same subject is repeated multiple times with the same stimuli. Therefore, calculating noise ceiling is not so difficult. Calculating correlation coefficients of each voxels between the first repeat and the second repeat will give me noise ceiling for the specific subject.

However, it is not always possible since some datasets do not have repeats. There are multiple subjects, but those subjects are just watching the movie once without any repeats. In this case, since there is no repeats, calculating noise ceiling is not straightforward. I wonder if there is any way to calculate noise ceiling in this case where there is no repeats for each subjects.

I am trying to predict voxel-wise activities from fMRI data, and noise ceiling is really needed to see if the prediction is meaningful of not.
If you have any advice, please share it with me.