AbstractPurposeThe aim of the study was to explore a novel methodology for designing universal pulses (UPs) that balances the benefits of a calibration‐free approach with subject‐specific online pulse design.MethodsThe proposed method involves segmenting the population into subpopulations with variability in anatomical shapes and positions reduced to 75%, 50%, and 25% of their original values while keeping the mean values unchanged. An additional 25% extreme case with a large volume of interest and shifted position was included. For each group, a 5kT‐points universal inversion pulse was designed and assessed by the normalized root mean square error (NRMSE) on the target longitudinal magnetization profile. The performance was compared to the conventional one‐size‐fits‐all approach. A total of 132 electromagnetic simulations were executed to generate representative anatomies and specific absorption rate (SAR) distributions in a three‐dimensional parameter space comprised of head breadth, head length, and Y‐shift. The 99.9^th percentile on the peak local SAR distribution was utilized to establish an intersubject variability safety margin.ResultsUPs designed for subpopulations with decreased head shape and position variability reduced the anatomical safety margin by up to 20%. Furthermore, when a head was significantly different to the average case, the proposed approach improved the inversion homogeneity by up to 24%, compared to the conventional one‐size‐fits‐all approach.ConclusionSubpopulation UPs present an opportunity to improve the homogeneity and reduce anatomical SAR safety margins at 7T without additional acquisition time for calibration.