The study investigated the potential for obtaining more accurate spine joint reaction force (JRF) estimates from musculoskeletal models by incorporating dynamic stereo X-ray imaging (DSX)-based in vivo lumbar vertebral rotational and translational kinematics compared to generic, rhythm (RHY)-based kinematics, while also observing the influence of accompanying inputs: intervertebral segment stiffness and neutral state. A full-body OpenSim® musculoskeletal model, constructed by combining existing lower- and upper-body models, was driven based on one volunteer’s (female; age 25; 60.8 kg; 176 cm) anthropometrics and kinematics from a series of upright standing and straight-legged dynamic lifting tasks. The lumbar spine portion was modified in a step-wise manner to observe effects of: (1) RHY vs. DSX lumbar kinematics; (2) No disc (bushing) stiffness (NBS); generic, linear bushing stiffness (LBS); subject-specific nonlinear bushing stiffness (NLBS); (3) Upright standing (UP) vs. Supine (SUP) neutral state; (4) Weight lifted: 4.5 kg vs. 13.6 kg. L4L5 JRF from 24 model variations based on combinations of aforementioned parameters were compared. Rhythm-based kinematics without translational components tends to over-predict JRF (31% and 39% for compression and shear, respectively) compared to DSX-based kinematics. Additionally, differences due to accompanying passive stiffness and neutral state choice combinations were even larger (>50%), indicating heightened demand on the quality of these accompanying inputs. The study not only highlights model sensitivity to choices made regarding the three primary inputs—kinematics, passive stiffness and neutral state— separately, but also how interactions between these choices can result in significant variability in joint loading estimates.