Medical image-based computational fluid dynamics (CFD) is a valuable tool for studying cardiovascular hemodynamics and its role in vascular pathologies. However, patient-specific flow rate measurements are rare. As a remedy, individual flow rates are typically estimated using anatomical features. In this study, we considered left atrial (LA) flow and compared three commonly used models to two proposed models based on mitral valve orifice area or atrial volume. We optimized our two models against the mean and standard deviation (SD) for heart rate (HR), cardiac output (CO), and peak velocities (E-wave, S-wave) using reference values from healthy individuals. All five models were evaluated on a cohort (N = 39), and the total error, calculated as the sum of the mean and SD, ranged from 37% to 71% for the commonly used models, while our proposed models achieved errors of 21.7% and 16.1%. For patients with atrial fibrillation (AF), we adapted our two models by removing the A-wave and minimizing relative differences in CO, E-wave velocity, and HR compared to normal sinus rhythm. The adapted models for AF showed errors of 1.3% and 1.1% when compared to the relative changes observed between AF and sinus rhythm patients. In conclusion, flow rates are one of the most sensitive yet influential parameters in cardiovascular simulations and can be standardized to improve model robustness. While fully patient-specific flow rate models are challenging to achieve, reducing variability against clinical measurements is a practical approach towards a plausible LA flow model.